SPECIAL SECTION ON ARTIFICIAL INTELLIGENCE (AI)-EMPOWERED

INTELLIGENT TRANSPORTATION SYSTEMS

Received December 8, 2019, accepted December 16, 2019, date of publication December 23, 2019, date of current version January 2, 2020.
Digital Object Identifier 10.1109/ACCESS.2019.2961372

Machine Learning Adoption in Blockchain-Based

Smart Applications: The Challenges,
and a Way Forward
SUDEEP TANWAR 1 , QASIM BHATIA 1 , PRUTHVI PATEL1 , APARNA KUMARI1 ,
PRADEEP KUMAR SINGH 2 , AND WEI-CHIANG HONG 3 , (Senior Member, IEEE)
1 Department of Computer Science and Engineering, Institute of Technology, Nirma University, Ahmedabad 382481, India
2 Department of Computer Science and Engineering, Jaypee University of Information Technology, Waknaghat 173234, India
3 Department of Information Management, Oriental Institute of Technology, New Taipei 220, Taiwan
Corresponding author: Wei-Chiang Hong ([email protected])
This work was supported by the Ministry of Science and Technology, Taiwan, under Grant MOST 108-2410-H-161-004.

ABSTRACT In recent years, the emergence of blockchain technology (BT) has become a unique, most
disruptive, and trending technology. The decentralized database in BT emphasizes data security and privacy.
Also, the consensus mechanism in it makes sure that data is secured and legitimate. Still, it raises new security
issues such as majority attack and double-spending. To handle the aforementioned issues, data analytics is
required on blockchain based secure data. Analytics on these data raises the importance of arisen technology
Machine Learning (ML). ML involves the rational amount of data to make precise decisions. Data reliability
and its sharing are very crucial in ML to improve the accuracy of results. The combination of these two
technologies (ML and BT) can provide highly precise results. In this paper, we present a detailed study on ML
adoption for making BT-based smart applications more resilient against attacks. There are various traditional
ML techniques, for instance, Support Vector Machines (SVM), clustering, bagging, and Deep Learning (DL)
algorithms such as Convolutional Neural Network (CNN) and Long short-term memory (LSTM) can be used
to analyse the attacks on a blockchain-based network. Further, we include how both the technologies can be
applied in several smart applications such as Unmanned Aerial Vehicle (UAV), Smart Grid (SG), healthcare,
and smart cities. Then, future research issues and challenges are explored. At last, a case study is presented
with a conclusion.

INDEX TERMS Blockchain, machine learning, smart grid, data security and privacy, data analytics, smart
applications.

I. INTRODUCTION makes the life of an individual easier. These applications gen-

From the past few decades, data has become an essential erate a huge amount of data, and storage of this ever-evolving
source of intelligence and carries new opportunities to the data in databases is a problem, and its communication also
real-life problems such as wireless communications, bioin- raises security concerns. To handle these issues, BT can
formatics [1], agriculture [2], and finance [3] through smart be used, which has a distributed database network. It was
applications. These applications are data-driven and incorpo- coined by Satoshi Nakamoto in the year 2008 and contained
rate actionable insights into user experience, which enables a time-stamped series of tamper-proof records, which are
individuals to complete the desired task more efficiently. managed by a cluster of distributed computers. It comprises
It operationalizes insights, personalizes the customer experi- of a chain of blocks that are connected using cryptographic
ence, optimizes customer interactions, improves operational primitives. The three mainstays of BT are immutability,
efficiency, and enable new business model. There are various decentralization, and transparency. These three characteris-
smart applications such as SG, UAV, Smart Cities, which tics opened its door for a wide variety of applications, for
example, digital currency existence (currency with no phys-
The associate editor coordinating the review of this manuscript and ical existence) and analysis on the suitability of it in smart
approving it for publication was Amr Tolba . applications [4]. Although BT ensures security and privacy

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see http://creativecommons.org/licenses/by/4.0/
474 VOLUME 8, 2020
S. Tanwar et al.: Machine Learning Adoption in Blockchain-Based Smart Applications

issues, some vulnerabilities also started appearing after its our knowledge, most of the surveys have focused on spe-
implementation. For instance, the nature of attacks began to cific areas, fields, or applications where it requires both ML
be increasingly sophisticated such as majority attacks (51% and blockchain. The proposed survey covers all the funda-
attack) that control voting, Sybil attacks for fake identity mental aspects of ML to be applied BT based applications,
generation to control the consensus [5]. To handle the afore- for instance, intrusion detection. A review conducted by
mentioned issue, a robust Intrusion Detection System (IDS) Meng et al. [12] describes how blockchain help to meet intru-
is required in place because the traditional methods use a sion detection. The authors mitigate trust issues by establish-
signature-based approach to detect specific patterns. But, ing a collaborative IDS using smart contracts [17]. Further,
to detect intrusions and attack patterns, one of the emerging Conti et al. [13] surveyed on security and privacy issues
technology known as ML can be used to analyze the data of bitcoin. They discuss various categories of attacks, such
traffic. Thus, designing efficient and effective algorithms to as double-spending attacks, client-side security threats, and
analyze this massive amount of data is in dire need of han- mining pool attacks. To handle the aforementioned issue,
dling the blockchain-based smart applications. Hence, ML is Rahouti et al. [5] discuss specific ML-based solutions and
highly prevalent today and uses a dozen times a day without discuss the handling of specific social issues such as human
even realizing it. ML encompasses computers to study, think, trafficking and drug sales through cryptocurrency. Then,
and act without intervention of humans. It is considered to Ucci et al [14] explored malware analysis using ML tech-
be one of the applications of Artificial Intelligence (AI). niques. Features of malware were thoroughly discussed, and
ML provides computers the competency to learn without a detailed taxonomy has been proposed. Salah et al. [15] and
being programmed it explicitly. Its basic idea is to build an Casino et al. [16] conducted a review on blockchain-based
efficient algorithm that can accept input data and, with the applications. To clarify the main difference between other
help of statistical analysis, make a prediction, and update the surveys paper and this survey paper, a comprehensive com-
outputs. A substantial amount of data can be analyzed by ML parison has been shown in Table 1. It includes objectives,
to create data-driven decisions. merits, and demerits of peer surveys concerning numerous
In a communication network of blockchain-based smart parameters such as architecture, applications, open issues,
applications, there is layer-wise handling of security issues. taxonomy, and merits, demerits of the existing approaches.
Some security issues are handled at the network layer, such as A master taxonomy of ML for BT is summarized in Figure 3.
malicious packets and some at the application layer such as
malware [6]. At the network layer, malicious packets can be B. RESEARCH CONTRIBUTIONS
used to impose the network to establish fraudulent consensus. Though several research works exist to address the ML
A naive solution to this problem can be to use a firewall usages for the blockchain-based applications, but not been
to ensure that packets meet pre-defined security criteria [7]. exploited to its full potential. In this paper, we investigated
Though, the attacks are becoming more sophisticated with the ML usages for blockchain based smart applications.
unseen patterns to bypass a firewall. To prevent this issue, Following are the research contribution of this paper.
packets header data can be analyzed using ML models [8] in
real-time using historical data. This analysis helps to detect • A brief discussion on how ML and blockchain can be
new and changing patterns. Similarly, ML techniques can used together in smart applications with a proposed
be used to classify malware to end-point such as servers, architecture.
mobile, or workstations. Further, several blockchain-based • To develop taxonomy covering ML techniques required
smart applications such as UAV [9], Data Trading [10], for BT based environment. In each part of the taxonomy,
SG build trust between data exchangers [11]. It is very crucial existing work has been discussed in detailed to handle
in any smart application at the same time; data should be several issues, such as preventing and predicting attacks
secure. BT ensures data security but to build confidence, on the blockchain network.
and ML techniques are used to predict untrustworthy nodes • A case study is presented to demonstrate the usage of
based on past patterns. Similarly, UAVs have significantly ML techniques in blockchain-based smart applications,
different network topology compared to the conventional such as SG, UAV, etc.
blockchain network topology [9]. It includes communication
using satellites and various ground stations. For UAV, BT is C. ORGANIZATION
used to securely store coordinates and other relevant data to The rest of the paper is organized as follows: Section 2 pro-
maintaining graph integrity for the vehicles. In subsequent vides a background of Blockchain, ML, and list all benefits
sections, we explore the recent research work on ML adoption while applying ML for BT based application. Then, we pro-
in the blockchain-based smart application. posed an architecture. Survey procedures and taxonomy pre-
sented in Section 3. Then, in Section 4-7, we present existing
A. SCOPE OF THIS SURVEY advancements in ML approaches for BT-based smart applica-
Several survey work has been published till date by the tions. A discussion on future research issues and challenges is
different researchers on the various aspects of the adoption presented in Section 8. Further, section 9 covers a case study
of ML in blockchain-based smart applications [5]. As per on the SG system, and finally, we concluded the paper.

VOLUME 8, 2020 475

 S. Tanwar et al.: Machine Learning Adoption in Blockchain-Based Smart Applications

TABLE 1. Comparison of the existing surveys with the proposed survey.

II. BACKGROUND interest [22]. It involves a mixture of supervised and unsuper-

A. MACHINE LEARNING vised ML techniques. Unsupervised learning may be applied
ML is the field of study that focuses on building applications to discover the structure of input variables, following which
that learn through experience. It is the ability to teach a it is used to make best guess predictions for the unlabeled
computer without programming it explicitly [18]. ML encom- data. It feeds that predicted data back into the supervised
passes its work from a diverse set of disciplines, includ- ML algorithm as training data and use the model to make
ing philosophy, information theory, probability and statistics, predictions on unseen data.
control theory, psychology and neurobiology, computational
B. BLOCKCHAIN
complexity, and artificial intelligence [19]. ML algorithms
are used in many application and benefited it as listed Blockchains are an immutable set of records that are cryp-
below: tographically linked together for audit [23]. It is similar to
an accounting ledger. Here, previous records in accounting
• In Data mining, large databases contain different pat- ledger cannot be changed, and new records need to be verified
terns that can be discovered automatically by using ML by a trusted party. The only difference between these two is
techniques to analyze outcomes, for instance, medical that new blocks (set of records) checked by a decentralized
treatments of a patient from health record databases or to structure of nodes that have a copy of the ledger. There
identify the creditworthiness of a person from financial is no centralized party to verify the records. Blockchain is
databases. formed by linking valid blocks together; the current block
• ML applies in areas where a deterministic algorithm contains the hash of the previous block, and so on, as shown in
is not promising, such as human face recognition from FIGURE 1. This makes blockchain traceable and resistant to
images. change [24]. Older blocks cannot be modified, in case they are
• Application domains where the adaptable programming changed in any way; their hash would change. This emphasis
is required, for instance, controlling manufacturing pro- to link hash in all subsequent blocks to make the blockchain
cesses as per the demand of the customer and adapting network valid again. A copy of the blockchain is available
to the varying reading interests of readers. with every individual within the network; henceforth, any
ML algorithm eare application specific and depends on changes can be cross verified by the other users. These copies
the output required by the system. There are several ML of the blockchain are updated with the addition of a new
algorithms, such as Supervised ML, Semi-Supervised ML, block. Then, everyone can see the block, depending on the
and Unsupervised ML. (i) Supervised ML uses statistical permissions assigned by the administrator. BT uses a crypto-
models to predict output in numerical data and classify the graphic secure hash algorithm (SHA) such as SHA-256 and
correct label [20]. Here, the most commonly known algo- SHA-512 to maintain the data integrity within the block.
rithms include the regression approach and decision trees. Each block has a unique hash value. For instance, Ethereum
(ii) Unsupervised ML does not have label data. Here, data uses Keccak-256 and Keccak-512, while Bitcoin uses double
samples are grouped into clusters depending on their sim- SHA-256. This SHA is a collision-resistant algorithm, where
ilarity or dissimilarity [21] using a different approach. For no two different input data could produce the same output
example, K means clustering and association rules algo- (hash value). Henceforward SHA can be used to check if the
rithms. (iii) Semi-Supervised ML is also a category of data is the same or not. There are various SHA algorithms

476 VOLUME 8, 2020

S. Tanwar et al.: Machine Learning Adoption in Blockchain-Based Smart Applications

FIGURE 1. Blockchain structure.

developed by the NSA and NIST and belongs to the SHA-2

family [25]. SHA was initially envisioned as a fragment of
the Digital Signature Standard (DSS) to produce the signa-
ture. Further, consensus algorithms are used to determine
block validity. The selection of algorithms depends on the
type of blockchain, such as public, private, and consortium
blockchain. The selected algorithm should ensure the con-
sensus among nodes [26]. It must be able to use resources
efficiently and tolerate a degree of safety during the event of FIGURE 2. Proposed architecture for Machine Learning adoption in
attacks. Blockchain-based Smart applications.
Further, a smart contract is a program (set of codes) that
runs on the blockchain and adds blocks whenever certain
conditions are met. It is defined by translating actual legal on the data, and hence data-related issues in ML models
contracts into programs to enforce the legal contract onto may be eliminated. ML-models can be based on specific
the blockchain [27]. It is similar to stored-procedures in segments of the chain rather than the entire data-set. This
relational databases. It is stored as scripts in the blockchain could give custom models for different applications, such as
and executed according to the data fed to them to pro- fraud detection and identity theft detection. Few of the benefit
duce outputs that are expected from the original con- are listed beneath when ML is applied:
tract [28]. It governs transactions either executed fully or • User authentication as a legitimate user for requesting or
partially based on the current input. The primary purpose performing any transaction in the blockchain network.
of a smart contract is to provide superior security with • BT provides a high level of security and trust.
a reduction in cost and delays associated with traditional • Blockchain integrates public ML models into smart con-
contracts. tracts to ensure that the conditions and terms which were
previously agreed are sustained.
C. INTEGRATION OF MACHINE LEARNING IN • BT helps in the reliable implementation of an incentive-
BLOCKCHAIN-BASED APPLICATIONS based system; thus, it encourages users/customers to
The learning capabilities of ML can be applied to blockchains contribute data. This huge data will help to improve ML
based applications to make them smarter. By using ML secu- model performance.
rity of the distributed ledger may be improved. ML may • ML models can be updated on-chain environment of BT
also be used to enhance the time taken to reach consensus with a small fee and off-chain, locally on an individual’s
by building better data sharing routes. Further, it creates an device without any costs
opportunity to build better models by taking advantage of the • Good data contributions can happen from users/
decentralized architecture of BT. We proposed architecture customers, these data consistently computed, and
for ML adoption in BT-based smart application, as shown rewards can be given to the users.
in Figure 2. Here, the smart application collects data from dif- • Tamper-proof smart contracts can be evaluated by differ-
ferent data sources such as sensors, smart devices, and Intenet ent machines (having different hardware configuration),
of Things (IoT) devices. Data collected from these devices ML models will not diverge from their potential and
get processed as part of smart applications. The blockchain produce results exactly as it is supposed to do.
work as an integral part of these smart applications. Then, • Payments processed in real-time with trust on a
ML can be applied to these application’s data for analysis blockchain environment.
(Data analytics and real-time analytics) and prediction. The • Blockchains tools, for instance, Ethereum, deal with
data sets used by ML models could be stored on a blockchain thousands of decentralized machines all over the world.
network. This reduces errors in the data such as duplication, This guaranteed users that it is never completely
missing data value, errors, and noise. Blockchains are focused unreachable or offline.

VOLUME 8, 2020 477

 S. Tanwar et al.: Machine Learning Adoption in Blockchain-Based Smart Applications

III. SURVEY PROCEDURES AND TAXONOMY classifiers run locally by the data owner, and the results
In this section, we refer to the procedures used to perform this shared with other users within the network. To handle the
study; for example, our search approach and inclusion criteria trust computation issue in CIDS, Alexopoulos et al. [29] pro-
for the final set of papers. Likewise, we present a detailed posed an approach to handle the alerts and data distribution
taxonomy based on our literature review. among the participants. Then, it verifies the data and exe-
cutes a suitable consensus algorithm to add a new data block
A. SYSTEMATIC LITERATURE REVIEW within the blockchain network. These data alerts could be
We used standard databases (for example IEEEXplore, ACM encrypted with keys distributed to selected parties. Another
Digital Library, ScienceDirect, and Springerlink) and Google approach was to keep them on a separate blockchain while
Scholar to search existing research work, using keywords still being part of the network. Further, an anomaly detection
such as ‘‘Machine Learning for Blockchain-based Smart system (ADS) was implemented based on the assumption
Application,’’ (‘‘Blockchain for Smart Applications’’ AND that a similar kind of attack may occur, but on different
‘‘Machine Learning for Smart applications’’). In the ini- nodes in the blockchain network [32]. This system does not
tial phase, publications emphasis essentially on blockchain discard information about orphans and forks that are usually
approaches for smart applications. Then, in the next phase done by other ADS. The attacked nodes shared this informa-
of the search, we concentrated on an ML-based solution for tion with other neighbor’s node within the network. In the
blockchain-based smart applications. Based on the results of experimental analysis, the system successfully prevented
these searches, we removed duplicate articles and obtained the same kind of attack (on other nodes) with negligible
our first set of more than 350 publications. Further, we steered overhead.
the search procedure with several magazines, journals, and
conferences dedicated to the parent field, ML, smart applica- B. MONITORING
tions, and Blockchain. Based on that, we found 130 articles.
A blockchain monitoring system has been prototyped using
Then, we studied the different sections of the articles like
Self Organising Maps (SOMs) [33]. It has modeled the
abstract, conclusion, and introduction. Then we categorized
dataset without external control. Here, large vectors map
these articles as ‘‘relevant’’ or non-relevant’’ to ML for
to smaller and lower dimension vectors using SOMs. The
blockchain-based smart applications. Lastly, we only selected
system analyzed blockchain data using Kohonen and SOM-
60 publications to present the taxonomy, as shown in Figure 3.
brero libraries (in the R programming environment). The
Here, each layer is color-coded to tailor each level of the
result shows the effective key attributes monitoring of the
taxonomy. For example, the root represents the level-0 of the
blockchain nodes and finds the emerging patterns. More,
classified taxonomy and presented in blue color. In a similar
another approach included the distributed pattern recogni-
way, the four major dimensions are representing as level-1 in
tion system with the concept of Graph Neurons (GNs) [31]
light-green color and so onwards.
to monitor the blockchain system. GNs are scalable and
could recognize patterns from similar or incomplete patterns.
B. TAXONOMY
The GN communicated with adjacent nodes to detect events
In this paper, we studied the papers that are predom-
within the network by using input data. Preliminary results
inantly based on either ML, blockchain, or both. The
showed object detection by the GN was accurate; still, further
presented survey has been divided into four dimensions:
work is needed in the area.
Goal Oriented, Layer Oriented, Countermeasures, and Smart
Applications. Figure 3 shows the taxonomy of ML adoption
for blockchain-based smart applications. C. PREDICTION
The ML models are mostly used for prediction. A good
IV. GOAL ORIENTED prediction model helps to make the right decision making
In this dimension, research work has been included from the and analytics. Added to this, an ML model to predict the
following aspects: (i) preventing, (ii) predicting, (iii) mon- bitcoin prices has been proposed by Valenkar et al. [30].
itoring, (iv) detection, and (v) response of blockchain. For This model uses bayesian regression and random forest
instance, [29] discussed the detection of intrusion in a collab- with several features such as block size, total bitcoins, day
orative manner, [30] predicts bitcoin prices using Bayesian high, number of transactions, and trade volume. The trained
regression method and [31] monitors the blockchain network dataset was normalized using log, z-score, and box-cox nor-
in a distributed manner. malization techniques. Further, a price prediction study has
been done for several cryptocurrencies such as Ripple, Lite-
A. DETECTION coin, Dash, Bitcoin, and Ethereum cryptocurrencies [34].
This subsection includes the detection of the attacks to handle Here, correlation matrices for feature selections used and
data security issues. Meng et al. [12] has proposed a Col- reported the general trends within the network. The pro-
laborative IDS (CIDS) with a data-sharing agreement on a posed model used multiple regression techniques on bit-
blockchain-based environment. Data privacy issues of CIDS coin. The system showed the 0.9944 accuracies for price
can be addressed with the use of ML classifiers on BT. These prediction.

478 VOLUME 8, 2020

S. Tanwar et al.: Machine Learning Adoption in Blockchain-Based Smart Applications

FIGURE 3. Proposed Taxonomy for Machine Learning adoption in Blockchain-based Smart Applications.

D. RESPONSE database of the previous organization. This ensures consis-

Tsolakis et al. [35] presented an approach for the secure tency and integrity of data as well as the authenticity of the
exchange of energy data between a cluster of end-users sender.
and a Virtual Node. This approach uses a blockchain-based A similar idea to tracking student’s learning history has
architecture to give a demand response solution. This solu- been proposed [40]. The main aim was to maintain the
tion included Fog-Enabled Intelligent Devices (FEID) [36] learning histories of the students and ensure access control,
at the user side (act as a blockchain node) and managed privacy, and security. To ensure privacy and access control,
smart contracts with the energy producer [36] in cloud three contracts were suggested: Learner - Learning Provider
computing platform [37], [38]. Then, a group of energy Contract (LLPC), Registrar - Learning Provider Contract
users combined into a centralized Virtual Node. All predic- (RLPC), and Index Contract (IC). The role of RLPC is to
tions and forecasting about the energy consumption have specify all conditions that describe request access by the
done on the Virtual Nodes. The end-user learned itself provider within the network. LLPC shows that a learner’s data
from each experience that helps to improve the accu- exists on the provider’s database. The IC map the learners
racy of data passed as the next iteration to the virtual and providers and the corresponding learning history. Here,
node. specialized Learning Blockchain Application Program Inter-
faces (LB APIs) has been proposed to encourage the adoption
E. PREVENTION of this technology. Initially, the system is started with a
One of the based uses of ML in BT based application is the boot node in the network. A new provider is added to the
prevention of an issue; for instance, employee verification by network as per the conditions described in RLPC. A new
a future employer. A mechanism proposed to reduce the time learner account is created and noted LLPC. In case, the new
taken to verify work history details provided by a prospective provider wants access to learners’ data, it has to request the
employee [39]. The previously worked organization has to learner using the LLPC, and upon granting, data will be
compile details of an employee (date of joining, leaving, post, accessible.
etc.), and a public key is generated to encrypt the data. Sub-
sequently, a new smart contract is created for the employee. F. COMPARISON OF EXISTING APPROACHES FOR GOAL
Then, the address of the smart contract is entered into the ORIENTED DIMENSION
organization’s database, and later it verified by the future A detailed comparison of the approaches discussed above
employer. In a future organization, work history is fetched is shown in Table 2. This comparison of works included
from the smart contract of employees and decrypted using several parameters such as trust management, ML, algorithm,
the key. Then, each work history entry compared with the blockchain, detection, merits, and demerits.

VOLUME 8, 2020 479

 S. Tanwar et al.: Machine Learning Adoption in Blockchain-Based Smart Applications

TABLE 2. Comparative analysis of goal oriented approaches.

V. LAYERS ORIENTED file to other nodes within the network. The file executed
At this moment, BT is very young, and understanding of locally in the local detection model of each node. The prob-
layer division is not enough. Considering the BT as a solitary ability value of being malware is appended with the block.
layer is similar to enduring everything among the physi- A weighted average for the trust of the node is applied to
cal layers and transport layers into a single layer. While the appended probabilities. The weighted average calculated
separating the BT into multiple layers, we can understand on the condition that the file is malware or not. The model
various properties of BT which are needed to be imple- used Restricted Boltzmann Machines (RBM) techniques with
mented. Some of the properties are: (i) Security: Nodes that 3000 hidden units. The results showed good accuracy for
do not control rare resources (commonly computing power) malware detection.
majorly cannot convince others for a different version of the
ledger. (ii) Liveness: Here, new blocks can be added to the B. ENDPOINT
blockchain with suitable latency. (iii) Stability: Nodes within Endpoints included the small computing devices which are
the blockchain network should not amend their belief of the participating in the BT based application network. To handle
consensus ledger except rare cases. (iv) Accuracy: Blocks the issue of fast computing and reduction in computation
added to the ledger must signify valid transactions such as power, emergent technology Edge Computing comes into the
they imitate to a description of how new blocks relate to picture. The Edge Computing Service Provider (ECSP) can
previous blocks. This dimension discusses the adoption of meet the requirements, and an approach has been proposed
ML techniques in various layers of a blockchain network, to maximize the revenue of ECSP [44]. The ECSP allocates
such as endpoints, application layer, process, network (intru- the resource unit to the highest bidder who participated in the
sion), and finally, user level. The proposed categorization into bidding process.
layers recognizes each property. Security is accomplished Similarly, Kim et al. [45] proposed an architecture to
at each layer, Liveness is realized at the user layer, Sta- offload resources computations in Deep Neural Networks
bility is achieved at the network and application layer, and to edge computing. Within this architecture, the embed-
accuracy is achieved at the endpoints, wherever blocks have ded device (the computation device at the edge server) and
significance. the edge server had to make initial deposits. After that,
the embedded device sends the computations to edge servers.
A. APPLICATION LAYER Then, the edge server returns the results, which are verified
The application layer helps to accomplish the security, live- by other nodes within the network. The server is rewarded
liness within the network. An approach to detect malware for its work after validation. The typical Ethereum architec-
(in the form of a portable executable file) has been proposed ture had been modified in this implementation by replacing
using DL methods [43]. The portable executable file was Ethereum virtual machine by V8 javascript virtual machine.
converted to a grayscale image to feed to the Deep belief The blocks generation part had been separated from the vir-
network. In this solution, the receiving node broadcast the tual machine to ensure that smart contract execution cannot

480 VOLUME 8, 2020

S. Tanwar et al.: Machine Learning Adoption in Blockchain-Based Smart Applications

TABLE 3. Comparative analysis for Layer-Oriented approaches.

affect the rate of generation of blocks. The system per- VI. COUNTERMEASURES
formance shows better results compare to the state-of-art The ML approached in response to a threat or monitoring
approaches. a system can be classified as real-time analytics or analysis
based on historical data [33]
C. PROCESS
This subsection comprises ML techniques to the process
A. ANALYTICAL
involves in the blockchain-based applications. An ML classi-
fier has been used to detect the human traffickers using adult In today’s world, every enterprise generates huge amounts
classified ads [46]. It included ads dataset from Backpage (a of data from different sources such as social media, smart-
website for online classifieds). It used a supervised learning phones, IoT, and other computing devices. These data are
model and used logistic regression for classification. The tremendously valuable to organizations. The overall tech-
labeled dataset with identifiers like email and phone number nique to find a meaningful pattern from these data is called
has been used for testing. This approach takes two ads at a data analytics. It is a process to convert data from foresight to
time as an input to the model. If the ads are from the same insight. It describes what happened in the past, draw aware-
author, the output will be same. The trained model showed an ness about the present, and make predictions (with some ML
89.54% true positive rate. Further, this approach used to build techniques) about the future. ML techniques can be categories
a graph and found links between ads and their related bit- as supervised, unsupervised, and reinforcement learning to
coin transactions. This approach showed a high false-positive analyses the data. The supervised learning included classifi-
rate. cation and regression originated on the idea of example-based
learning. Next, unsupervised learning techniques perform
D. USER clustering, dimensionality reduction, and recommendation
Users are the individual or system that uses the sys- of a system based on the dataset. This used to recognize
tem functionality seating at one end of the application. hidden patterns or focus on the well-educated behavior of
Wasim et. al [47] proposed law as a service ML-based archi- the machine. The reinforcement learning approach helps to
tecture to monitor contract breaches. Previously, contract reward maximization. In the next decades, our society will
breaches were dealt with or without the use of courts. The be driven by new technological developments in ML and
proposed system used an unsupervised ML algorithm called BT. Further, a traceability algorithm has been proposed for
Probability-based Factor Model to issue injunctions. This bitcoin mining by using the ANN approach [51]. It aimed to
model simulated using three service providers Redis [48], remove irrelevant data in mining and introduce traceability to
MongoDB [49], and Memcached Servers [50]. Services mon- the system. This was helpful in a distributed architecture and
itor the contracts for breaches. The results showed that ser- decrease traceability time.
vices that perform complex operations are more likely to DL is a subset of ML that is originated on definitive
breach contracts. algorithms influenced by the overall structure and work of the
neural network in the human brain. DL rationalizes tasks to
E. COMPARISON OF EXISTING APPROACHES FOR LAYER perform speech recognition, image recognition, make insight-
ORIENTED USE CASES ful decisions on natural language processing. DL network
A detailed comparison of the approaches is discussed consisted of input, output, and hidden multi-layer. It accepted
in Table 3. This comparison is made based on several a new block and a history of the previous block as an input
parameters such as edge computing, blockchain, neural net- in BT based applications. It used a state-transition algorithm
work, supervised learning, pros, and cons of the proposed on the features such as hash value, nonce, address, and trans-
approach. action data. An ML-based classification approach is used to

VOLUME 8, 2020 481

 S. Tanwar et al.: Machine Learning Adoption in Blockchain-Based Smart Applications

identify the cyber-crime in Bitcoin [52]. The results provided

a glimpse of the size of cybercrime in the Bitcoin ecosystem.

B. REAL-TIME
ML enables real-time analytics for all types of data, such
as social - accessible, transactional, and operational. It uses
in-memory computing though leaving unremittingly updated
data securely. It improves analytics accuracy and acceler-
ates the predictive behavior of ML models. It has been
noticed that BT makes real-time cross edge transactions in
monetary and payment frameworks. Several fintech innova-
tors and banks are currently investigating blockchain due to
fast and real-time settlement of massive amounts indepen-
dent of geographic hindrances. Similarly, associations that
require real-time analytics of information on a huge scale
can approach an ML and blockchain-empowered framework
to accomplish the goal. With ML and BT, financial insti-
tutions and other associated organizations across the globe FIGURE 4. A list of smart applications.

can trace the data changes to make quick and fast business
decisions regardless of irregular activities or suspicious trans-
actions. Moreover, the performance of the blockchain-based such as shops, malls, online stores), (ii) Customer Layer,
Software-Defined Vehicular Network (SDVN) has been (iii) Application Layer (applications provide services to the
improved using deep Q-learning methodology [53], [54]. consumers), (iv) Data conversion layer(Data masking and
This approach used a permissioned blockchain with Byzan- unmasking), (v) AutoML layer (Consists various ML models
tine fault tolerance as a consensus protocol. Results showed such as Linear regression, SVM and logistic regression for
that this scheme managed network and computing resources predictions), (vi) Blockchain Layer (Secure storage of data
better and gave the best throughput in the SDVN [55]. Fur- and ensures the safe data exchange) This framework helps
ther, Liu et al. [22] recommended a data collection frame- organizations to keep their data safe, automate their processes
work for Industrial-IoT (IIoT) applications. This framework and share data with other organizations in a mutually benefi-
combines the use of deep reinforcement learning (DRL) and cial and safe way.
Ethereum blockchain. To store and share data, Ethereum
nodes were categories into two categories: mining nodes B. DEVICE PERSONALIZATION
and nonmining nodes. The proposed DRL algorithm used Device Personalization (DP) is a component that uses pre-
three plain components, actions (moving path and remoteness dictions across smart devices to improves the quality of
of mobile terminals), states (environment description), and service (QoS) such as actions in the launcher, smart text
rewards (amount achieved). The proposed algorithm shows selection during writing on text pad. DP Services uses system
the 34.5% increase in geographical fairness compared to a permissions to provide smart predictions. In a smart home
random solution. environment, and ML model-based single DP framework has
been implemented [59]. Here, smart devices connected to a
VII. SMART APPLICATIONS smart hub. Every time a user uses a device, a log is generated
The last dimension covers the adoption of ML techniques with user data, device data, and other parameters for that
to BT-based smart applications such as data trading, UAV, device. For example, uses of air conditioner at home. This
product manufacturing, Medical, and Healthcare [56], Smart log data help to adjust the operation of the device.
Cities, automated Customer Service, and Device Personaliza-
tion as shown in Figure 4. ML and BT are revolutionizing up- C. PRODUCT MANUFACTURING
to-date technologies by transforming customer experiences, As a feature of the manufacturing process, organizations have
behaviors, and business models [57]. Both are making strides started trusting blockchain-based procedures to empower
in the major smart application: production, security, transparency, and compliance checks.
Instead of following fixed schedules of machine maintenance,
A. CUSTOMER SERVICE ML algorithms are being used to make flexible plans at
With the upsurge of the customer, the Customer service has specific periods. Product testing and quality control have
to be more efficient and automated to meet rising customer also automated increasingly, with versatile algorithms. It suc-
needs. One of the best ways is to automate the process to cessfully detects faulty and good products, particularly in
increase a company’s capabilities. Wang et al. [58] proposed profoundly delicate situations. For example, Porsche (a car
an AutoML framework for the blockchain-based application. manufacturing company) is one of those early adopters of
It consists six layers: (i) Organization Layer (Includes entities ML and BT technologies; to improve automobile safety and

482 VOLUME 8, 2020

S. Tanwar et al.: Machine Learning Adoption in Blockchain-Based Smart Applications

increase capabilities. The organization utilizes blockchain parking lots must be recognized spontaneously to benefits in
innovation to transfer data more safely and rapidly, offering dealers and vehicle owner’s daily life. (iii) Smart Weather and
its clients peace of mind, regardless of parking, charging, and Water Systems: Here, the system can use some sensors to
third-party access to their car. generate appropriate data such as temperature, wind speed,
rain, and pressure. The analysis of these data through ML
D. DATA TRADING techniques can contribute to improving the density of smart
DataTrading is an innovative platform that makes advanced cities. (iv) Smart Vehicular Traffic: Vehicular traffic data with
trading possible for retail traders from all over the world. a suitable analysis will benefit the government and citizens
An Ethereum based data trading framework has presented to a great extent [62]. Everyone can decide the arrival time to
that succeeded in preventing single-point failure and preserv- a destination by using these data. (v) Surveillance Systems:
ing privacy at the same time [10]. The framework consists of Physical security is an utmost important concern for citizens
three entities, data provider, a data consumer, and a market anywhere they live. To address this issue, smart technology
manager. Once the network is set up, the data consumer and such as ML and BT can be configured for it. Consequently,
data provider registered themselves with the market manager. collecting and analyzing data and identifying crimes is one of
The data provider needed to deposit an amount with the man- the challenging tasks. (vi) Smart Healthcare: It includes the
ager greater than the amount to be paid by the consumer. A list accessibility of the caregivers and doctors, identification of
indicating topics of data available with the provider is pub- nearest medical stores, and clinic. Hence, ML and BT play
lished. The client referred it and requests some encrypted data a vital role while data is getting generated [63], [64]. (vii)
blocks for content validation using a distance metric learning Smart Governance and Smart Education: Smart governance
technique. After successful validation, the client responded, can maintain a city smartly. A smart city includes a differ-
and the provider sent a signature, and the client published a ent way of education. It captures the data of the students
smart contract to the network. The provider then sent another and employees in educational and government institutions.
signature, and these two signatures, the client, decrypted the Predictions and analytics required to keep the stuff up to
data. There are different protocols such as setup protocol, standards.
register protocol, payment and acknowledgment protocol,
and query protocol that facilitate the communication between G. ENERGY AND UTILITIES
provider, consumer, and the manager. In the Energy industry, BT is assisting to simplify energy
exchanges. For example, IOTA enables smart transformation
E. UNMANNED AERIAL VEHICLE across the entire energy industry by implementing BT [65].
A UAV or drone is an aircraft that runs without a It uses the concept of peer-to-peer (P2P) energy production
pilot (human) aboard. Kuzmin et al. [9] has proposed and consumption. Smart energy microgrids are progres-
blockchain-based UAVNet model includes different devices sively developing fame as a technique of making ecological
such as a network of satellites, cellular base stations, and energy resources. LO3 Energy (a Newyork-based organiza-
ground control stations. Here, BT served multiple purposes, tion) is also using a blockchain-based development to develop
such as preserving the integrity of the data and for distributed energy generation, transmission, conservation, and exchang-
graphs computation. The communication between satellites ing within neighborhood networks. The technology uses
and base stations is prone to electromagnetic jamming, hence microgrid and smart meters, together with smart contracts,
a blockchain-based system enables UAVs to store relevant to manage and track energy transactions. GE Digital, together
coordinate’s data and operate autonomously within the jam- with Evolution Energie (a startup), has created an application
ming zone. This solution used a proof-of-graph consensus that encourages the tracking of renewable energy in SG and
algorithm with a simplified memory bounded algorithm of uses blockchain to give energy sources certificate. The idea
any existing shortest path to validating a new transaction. is allowing organizations and individuals to trade renewable
energy sources without the involvement of third-party [36].
F. SMART CITIES Globally, most of the industries are dogged by third-party,
Smart cities improve the living experiences of individuals. who increase the business cost. BT has already interrupted
ML and blockchain emergent technologies play a crucial role that model by facilitating the P2P model for customers. The
in the innovation of smart cities to provide critical services revolution works even better when combined with the ML and
and infrastructure components such as healthcare [60], city BT together. We have been witnessed the financial revolution
administration, smart homes, education, transportation, real by using the cryptocurrency world; similarly, we are going to
estate, and utilities [61]. (i) Smart Homes: Smart homes observe more innovation disruptions as a result of combining
can be monitored, and DP can help to improve the quality ML and BT.
of livelihood. (ii) Smart Parking System: Here, arrival and
departure of vehicles can be tracked for different parking lots VIII. FUTURE RESEARCH ISSUES AND CHALLENGES
available in a decentralized manner within the smart city [61]. Researchers are looking forward to these technologies across
Consequently, smart parking lots should be designed by see- the globe, but still, various obstacles resist the integration
ing the number of cars in each region. Furthermore, new of BT and ML [66]. Their integration is still in its infancy.

VOLUME 8, 2020 483

 S. Tanwar et al.: Machine Learning Adoption in Blockchain-Based Smart Applications

by all nodes, which creates a significant memory constraint

on these devices. The performance is also affected by an
increase in chain size. Besides that, the storage of irrelevant
data also wastes computational resources. Blockchains are
immutable, and hence storage management is a major issue
in most implementations.

E. IMPLEMENTATION
A large- scale blockchain network will require an equiva-
lently large number of transactions. Implementation of this
size of blockchain invites potential issues; for instance, high
demand for internet bandwidth, which not easy to reduce,
transactions will be a burden on the network. Hence, the addi-
tion of blocks and transactions needs to be decreased to meet
the inevitable demand.

F. SECURITY
Blockchains are decentralized, and they are prone to security
issues [68]. The most common concern is that due to attacks,
FIGURE 5. Future research issues and challenges. the consensus protocol may be compromised, such that the
mining power of a few farms will control which blocks are
Many open issues and challenges are yet to be addressed. added to the network. This particular concern is present in
Here, we discuss the futuristic open issues and challenges public blockchains. Private versions are unaffected by this
of ML adoption in BT for secure communication, as shown attack as they have each node identified, and an appropriate
in Figure 5. We highlighted the challenges as Suitability, consensus protocol is in place.
Infrastructure, Privacy, security, Memory, Implementation,
and Quantum resilience. G. QUANTUM RESILIENCE
The hashing algorithms used by blockchains could soon be
A. SUITABILITY broken with quantum computers. A blockchain is predom-
Blockchain is a viable solution if the source of data cannot be inantly at risk from this because it uses one-way functions
trusted, and several entities are high in the distributed envi- for encryption (only protection in digital signature). This
ronment. If performance is required, then a simple database would cause all the features that make blockchains a viable
is a better option. Therefore, the architecture of blockchain storage structure (obsolete). Luckily, quantum computing
must be understood before its use in any application [67]. offers opportunities to boost the performance and security of
blockchains. Quantum cryptography can reinforce the secu-
B. INFRASTRUCTURE rity of the blockchain network as quantum communication is
Blockchain specific hardware and network infrastructure authenticated inherently (users cannot mimic another user).
improve the performance of many blockchain-based appli- It can encrypt entire P2P communications and replace clas-
cations. These could include network administration, mining sical digital signatures in the blockchain network. Research
hardware, decentralized storage, and communication proto- is underway to design blockchain with quantum comput-
cols [30]. However, products tailored to use in blockchain are ing [69].
still under investigation (involving major tech companies and
financial institutions). IX. CASE STUDY
In order to demonstrate the proposed architecture for ML
C. PRIVACY adoption in blockchain-based smart applications, we present
Data generated by devices to be stored on the blockchain is a case study, as shown in Figure 6. In this case study, we study
available to the entire blockchain nodes [33]. This leads to a a blockchain-based SG system for energy transactions using
potential privacy concern for data that needs to be kept either cryptocurrency [70], [71]. This framework uses a blockchain
private or confidential. Such issues could be resolved by the and DL approach to complete an energy transaction. It is
use of private blockchains, controlled access, and encryption. a reliable P2P energy system. It is based on the Byzantine
However, the ML models adoption on these limited data fault tolerance algorithm to produce high system through-
imposed barriers for predictions and analytics. put. This approach consists of five phases (i)setup phase,
(ii) agreement phase, (iii) consensus-making phase, (iv) block
D. MEMORY creation phase, and (v) a change view phase. Here, blocks
The size of blockchain keeps growing as new blocks are are generated using hash functions and a short signature. It is
added to it. Consequently, the entire chain must be stored an IDS that works on recurrent neural networks to detect

484 VOLUME 8, 2020

S. Tanwar et al.: Machine Learning Adoption in Blockchain-Based Smart Applications

FIGURE 6. An energy trading system.

fraudulent transactions and network attacks in blockchain- network. After validation of the buyer on the blockchain
based energy applications. The performance of this IDS has network seller sells the energy to the buyer and receives
been studied on different energy datasets. It consists of four cryptocurrency in his account. This P2P energy trading sys-
entities for communications: (a) Energy Buyer, (b) Energy tem required further data analysis to identify the frequent
Seller, (c) Blockchain, and (d) IDS [41]. Energy buyer buyer and seller for the recognition of malicious transactions
trade energy with the energy seller. An energy buyer shows within the network. A malicious block or node can impact
that he has sufficient energy money (cryptocurrency) that the P2P system and compete for energy trading. IDS can help
fulfills the minimum asset requirement of energy sellers. to figure out the deceitful transactions and network attacks
An energy buyer will be a person or commercial building on energy trading applications [72]. The result shows the
or industry. Energy buyers can be located in a Home Area good performance of the system compares to the state-of-art
Network (HAN), Building Area Network (BAN), or Neigh- approaches.
borhood area Network (NAN). Energy seller entities demon-
strate that it has sufficient energy to sell. Energy sellers can be X. CONCLUSION
located in HAN, BAN, NAN, or energy companies. Energy The recent advancements in Blockchain and ML have made
seller (entities) produces energy from renewable sources such them path-breaking technologies. The distributed ledger has
as wind energy, solar energy, and biomass. The seller can be the possibility to work as the backbone of various smart appli-
a neighbor, local society with renewable energy resources, cations such as smart cities, UAV, SG, data trading. In this
utility provider, or SG. Blockchain entity is a distributed paper, we have presented detailed information on BT and ML,
digital ledger, which is encompassing all energy transactions along with their usages in smart applications and proposed
in the SG system. an ML-BT based architecture. This architecture can be used
Once an energy seller produces energy from renewable to design and deploy an ML-BT based data analysis system.
energy sources and uses that energy for his own use. After A discussion and comparison of various existing surveys are
that, if the seller left with energy, then he publishes his presented. Then, we presented ML-BT solution taxonomy,
per-unit prices per kilowatt energy on the blockchain net- focusing on goal oriented, layer oriented, countermeasures,
work. In this P2P energy trading system, the needed energy and smart application dimensions. A comparative analysis
buyer will look upon the published price and unit of energy. of available methodologies and approaches is presented in
First of all, the energy buyer checks his account that he has each dimension. Then, we have listed several research chal-
sufficient cryptocurrency balance for energy trading. Then, lenges being faced during ML adoption in BT-based systems,
if the energy demand of energy buyers and per-unit price which require solutions. We also emphasized a number of
matched his requirement of energy at that moment, he sends research prospects such as infrastructure availability, quan-
a purchase request to the energy seller through the blockchain tum resilience, and privacy issues that can serve as future

VOLUME 8, 2020 485

 S. Tanwar et al.: Machine Learning Adoption in Blockchain-Based Smart Applications

research directions in this field. Then, we presented a case [23] Z. Zheng, S. Xie, H. Dai, and H. Wang, ‘‘An overview of blockchain
study on the energy trading system to verify the effectiveness technology: Architecture, consensus, and future trends,’’ in Proc. IEEE Int.
Congr. Big Data, Big Data Congr., Jun. 2017, pp. 557–564.
of the proposed architecture and concluded the paper at last. [24] G. B. Mermer, E. Zeydan, and S. S. Arslan, ‘‘An overview of blockchain
technologies: Principles, opportunities and challenges,’’ in Proc. 26th Sig-
REFERENCES nal Process. Commun. Appl. Conf. (SIU), May 2018, pp. 1–4.
[25] W. Penard and T. van Werkhoven, ‘‘On the secure hash algorithm family,’’
[1] S. Kaneriya, J. Vora, S. Tanwar, and S. Tyagi, ‘‘Standardising the use of Cryptogr. Context, pp. 1–18, 2008.
duplex channels in 5G-WiFi networking for ambient assisted living,’’ in [26] M. Vukolić, ‘‘The quest for scalable blockchain fabric: Proof-of-work vs.
Proc. IEEE Int. Conf. Commun. Workshops (ICC Workshops), May 2019, BFT replication,’’ in Open Problems in Network Security, J. Camenisch
pp. 1–6. and D. Kesdoǧan, Eds. Cham, Switzerland: Springer, 2016, pp. 112–125.
[2] S. Tyagi, M. S. Obaidat, S. Tanwar, N. Kumar, and M. Lal, ‘‘Sensor cloud
[27] S. Thompson, P. L. Seijas, and D. Adams, ‘‘Scripting smart contracts for
based measurement to management system for precise irrigation,’’ in Proc.
distributed ledger technology,’’ Tech. Rep., Dec. 2016. [Online]. Available:
IEEE Global Commun. Conf. (GLOBECOM), Dec. 2017, pp. 1–6.
https://kar.kent.ac.uk/61162/
[3] U. Bodkhe, P. Bhattacharya, S. Tanwar, S. Tyagi, N. Kumar, and
[28] K. Christidis and M. Devetsikiotis, ‘‘Blockchains and smart contracts for
M. S. Obaidat, ‘‘Blohost: Blockchain enabled smart tourism and hospitality
the Internet of Things,’’ IEEE Access, vol. 4, pp. 2292–2303, 2016.
management,’’ in Proc. Int. Conf. Comput., Inf. Telecommun. Syst. (CITS),
Aug. 2019, pp. 1–5. [29] N. Alexopoulos, E. Vasilomanolakis, N. R. Ivánkó, and M. Mühlhäuser,
[4] S. Nakamoto, Bitcoin: A Peer-to-Peer Electronic Cash System, 2009. ‘‘Towards blockchain-based collaborative intrusion detection systems,’’
[Online]. Available: https://bitcoin.org/en/bitcoin-paper in Critical Information Infrastructures Security, G. D’Agostino and
[5] M. Rahouti, K. Xiong, and N. Ghani, ‘‘Bitcoin concepts, threats, A. Scala, Eds. Cham, Switzerland: Springer, 2018, pp. 107–118.
and machine-learning security solutions,’’ IEEE Access, vol. 6, [30] S. Velankar, S. Valecha, and S. Maji, ‘‘Bitcoin price prediction using
pp. 67189–67205, 2018. machine learning,’’ in Proc. 20th Int. Conf. Adv. Commun. Tech-
[6] A. P. Namanya, A. Cullen, I. U. Awan, and J. P. Disso, ‘‘The world of nol. (ICACT), Feb. 2018, pp. 144–147.
malware: An overview,’’ in Proc. IEEE 6th Int. Conf. Future Internet [31] A. Hudaya, M. Amin, N. M. Ahmad, and S. Kannan, ‘‘Integrating dis-
Things Cloud (FiCloud), Aug. 2018, pp. 420–427. tributed pattern recognition technique for event monitoring within the
[7] S. Krit and E. Haimoud, ‘‘Overview of firewalls: Types and policies: iot-blockchain network,’’ in Proc. Int. Conf. Intell. Adv. Syst. (ICIAS),
Managing windows embedded firewall programmatically,’’ in Proc. Int. Aug. 2018, pp. 1–6.
Conf. Eng. MIS (ICEMIS), May 2017, pp. 1–7. [32] M. Signorini, M. Pontecorvi, W. Kanoun, and R. Di Pietro,
[8] G. Betarte, E. Gimenez, R. Martinez, and A. Pardo, ‘‘Improving Web ‘‘BAD: Blockchain anomaly detection,’’ 2018, arXiv:1807.03833.
application firewalls through anomaly detection,’’ in Proc. 17th IEEE Int. [Online]. Available: https://arxiv.org/abs/1807.03833
Conf. Mach. Learn. Appl. (ICMLA), Dec. 2018, pp. 779–784. [33] S. Chawathe, ‘‘Monitoring blockchains with self-organizing maps,’’ in
[9] A. Kuzmin and E. Znak, ‘‘Blockchain-base structures for a secure and Proc. 17th IEEE Int. Conf. Trust, Secur. Privacy Comput. Commun./12th
operate network of semi-autonomous unmanned aerial vehicles,’’ in Proc. IEEE Int. Conf. Big Data Sci. Eng. (TrustCom/BigDataSE), Aug. 2018,
IEEE Int. Conf. Service Oper. Logistics, Informat. (SOLI), Jul. 2018, pp. 1870–1875.
pp. 32–37. [34] M. Saad and A. Mohaisen, ‘‘Towards characterizing blockchain-based
[10] Y. Zhao, Y. Yu, Y. Li, G. Han, and X. Du, ‘‘Machine learning based cryptocurrencies for highly-accurate predictions,’’ in Proc. IEEE Conf.
privacy-preserving fair data trading in big data market,’’ Inf. Sci., vol. 478, Comput. Commun. Workshops (INFOCOM WKSHPS), Apr. 2018,
pp. 449–460, Apr. 2019. pp. 704–709.
[11] C. Pop, T. Cioara, M. Antal, I. Anghel, I. Salomie, and M. Bertoncini, [35] A. C. Tsolakis, I. Moschos, K. Votis, D. Ioannidis, T. Dimitrios, P. Pandey,
‘‘Blockchain based decentralized management of demand response pro- S. Katsikas, E. Kotsakis, and R. García-Castro, ‘‘A secured and trusted
grams in smart energy grids,’’ Sensors, vol. 18, no. 1, p. 162, Jan. 2018. demand response system based on blockchain technologies,’’ in Proc.
[12] W. Meng, E. W. Tischhauser, Q. Wang, Y. Wang, and J. Han, ‘‘When Innov. Intell. Syst. Appl. (INISTA), Jul. 2018, pp. 1–6.
intrusion detection meets blockchain technology: A review,’’ IEEE Access, [36] A. Kumari, S. Tanwar, S. Tyagi, N. Kumar, M. S. Obaidat, and
vol. 6, pp. 10179–10188, 2018. J. J. Rodrigues, ‘‘Fog computing for smart grid systems in the 5G environ-
[13] M. Conti, E. S. Kumar, C. Lal, and S. Ruj, ‘‘A survey on security and ment: Challenges and solutions,’’ IEEE Wireless Commun., vol. 26, no. 3,
privacy issues of bitcoin,’’ IEEE Commun. Surveys Tuts., vol. 20, no. 4, pp. 47–53, Jun. 2019.
pp. 3416–3452, 4th Quart., 2018. [37] J. Vora, P. DevMurari, S. Tanwar, S. Tyagi, N. Kumar, and M. S. Obaidat,
[14] D. Ucci, L. Aniello, and R. Baldoni, ‘‘Survey of machine learning tech- ‘‘Blind signatures based secured e-healthcare system,’’ in Proc. Int. Conf.
niques for malware analysis,’’ Comput. Secur., vol. 81, pp. 123–147, Comput., Inf. Telecommun. Syst. (CITS), Jul. 2018, pp. 1–5.
Mar. 2019.
[38] S. Tanwar, J. Vora, S. Kaneriya, and S. Tyagi, ‘‘Fog-based enhanced safety
[15] K. Salah, M. Rehman, N. Nizamuddin, and A. Al-Fuqaha, ‘‘Blockchain
management system for miners,’’ in Proc. 3rd Int. Conf. Adv. Comput.,
for AI: Review and open research challenges,’’ IEEE Access, vol. 7,
Commun. Autom. (ICACCA), Sep. 2017, pp. 1–6.
pp. 10127–10149, 2019.
[39] P. Sarda, M. J. M. Chowdhury, A. Colman, M. A. Kabir, and J. Han,
[16] F. Casino, T. Dasaklis, and C. Patsakis, ‘‘A systematic literature review
‘‘Blockchain for fraud prevention: A work-history fraud prevention sys-
of blockchain-based applications: Current status, classification and open
tem,’’ in Proc. 17th IEEE Int. Conf. Trust, Secur. Privacy Comput. Com-
issues,’’ Telematics Inform., vol. 36, pp. 55–81, Mar. 2018.
mun./12th IEEE Int. Conf. Big Data Sci. Eng. (TrustCom/BigDataSE),
[17] W. Xiong and L. Xiong, ‘‘Smart contract based data trading mode
Aug. 2018, pp. 1858–1863.
using blockchain and machine learning,’’ IEEE Access, vol. 7,
pp. 102331–102344, 2019. [40] P. Ocheja, B. Flanagan, and H. Ogata, ‘‘Connecting decentralized learning
[18] T. M. Mitchell, Machine Learning, 1 ed. New York, NY, USA: records: A blockchain based learning analytics platform,’’ in Proc. 8th Int.
McGraw-Hill, 1997. Conf. Learn. Anal. Knowl., New York, NY, USA, 2018, pp. 265–269.
[19] P. Louridas and C. Ebert, ‘‘Machine learning,’’ IEEE Softw., vol. 33, no. 5, [41] H. Liang, J. Wu, S. Mumtaz, J. Li, X. Lin, and M. Wen, ‘‘MBID: Micro-
pp. 110–115, May 2016. blockchain-based geographical dynamic intrusion detection for V2X,’’
[20] R. Saravanan and P. Sujatha, ‘‘A state of art techniques on machine learning IEEE Commun. Mag., vol. 57, no. 10, pp. 77–83, Oct. 2019.
algorithms: A perspective of supervised learning approaches in data clas- [42] K. Sgantzos and I. Grigg, ‘‘Artificial intelligence implementations on the
sification,’’ in Proc. 2nd Int. Conf. Intell. Comput. Control Syst. (ICICCS), blockchain. Use cases and future applications,’’ Future Internet, vol. 11,
Jun. 2018, pp. 945–949. no. 8, p. 170, 2019.
[21] V. Vats, L. Zhang, S. Chatterjee, S. Ahmed, E. Enziama, and K. Tepe, [43] S. Raje, S. Vaderia, N. Wilson, and R. Panigrahi, ‘‘Decentralised firewall
‘‘A comparative analysis of unsupervised machine techniques for liver for malware detection,’’ in Proc. Int. Conf. Adv. Comput., Commun. Con-
disease prediction,’’ in Proc. IEEE Int. Symp. Signal Process. Inf. Tech- trol (ICAC), Dec. 2017, pp. 1–5.
nol. (ISSPIT), Dec. 2018, pp. 486–489. [44] N. C. Luong, Z. Xiong, P. Wang, and D. Niyato, ‘‘Optimal auction for edge
[22] C. Liu, X. Xu, and D. Hu, ‘‘Multiobjective reinforcement learning: A com- computing resource management in mobile blockchain networks: A deep
prehensive overview,’’ IEEE Trans. Syst., Man, Cybern, Syst., vol. 45, no. 3, learning approach,’’ in Proc. IEEE Int. Conf. Commun. (ICC), May 2018,
pp. 385–398, Mar. 2015. pp. 1–6.

486 VOLUME 8, 2020

S. Tanwar et al.: Machine Learning Adoption in Blockchain-Based Smart Applications

[45] J.-Y. Kim and S.-M. Moon, ‘‘Blockchain-based edge computing for deep [68] A. Kumari, S. Tanwar, S. Tyagi, and N. Kumar, ‘‘Verification and vali-
neural network applications,’’ in Proc. Workshop Intell. Embedded Syst. dation techniques for streaming big data analytics in Internet of Things
Archit. Appl. (INTESA), New York, NY, USA, 2018, pp. 53–55. environment,’’ IET Netw., vol. 8, no. 2, pp. 92–100, 2018.
[46] R. S. Portnoff, D. Y. Huang, P. Doerfler, S. Afroz, and D. McCoy, ‘‘Back- [69] A. K. Fedorov, E. O. Kiktenko, and A. I. Lvovsky, ‘‘Quantum computers
page and bitcoin: Uncovering human traffickers,’’ in Proc. 23rd ACM put blockchain security at risk,’’ Nature, vol. 563, pp. 465–467, Nov. 2018.
SIGKDD Int. Conf. Knowl. Discovery Data Mining (KDD), New York, NY, [70] M. A. Ferrag and L. Maglaras, ‘‘DeepCoin: A novel deep learning and
USA, 2017, pp. 1595–1604. blockchain-based energy exchange framework for smart grids,’’ IEEE
[47] M. U. Wasim, A. A. Z. A. Ibrahim, P. Bouvry, and T. Limba, ‘‘Law as a Trans. Eng. Manage., to be published.
service (LAAS): Enabling legal protection over a blockchain network,’’ in [71] N. Wang, X. Zhou, X. Lu, Z. Guan, L. Wu, X. Du, and M. Guizani, ‘‘When
Proc. 14th Int. Conf. Smart Cities, Improving Qual. Life Using ICT IoT energy trading meets blockchain in electrical power system: The state of
(HONET-ICT), Oct. 2017, pp. 110–114. the art,’’ Appl. Sci., vol. 9, no. 8, p. 1561, Apr. 2019.
[48] Redis. Accessed: Dec. 2019. [Online]. Available: https://redis.io/ [72] N. Kabra, P. Bhattacharya, S. Tanwar, and S. Tyagi, ‘‘Mudrachain:
[49] Mongodb. Accessed: Nov. 4, 2019. [Online]. Available: https://www. Blockchain-based framework for automated cheque clearance in financial
mongodb.com/ institutions,’’ Future Gener. Comput. Syst., vol. 102, pp. 574–587, 2020.
[50] Memcached. Accessed: Nov. 11, 2019. [Online]. Available: https://
memcached.org/
[51] R.-Y. Chen, ‘‘A traceability chain algorithm for artificial neural networks SUDEEP TANWAR received the B.Tech. degree
using T–S fuzzy cognitive maps in blockchain,’’ Future Gener. Comput. from Kurukshetra University, India, in 2002,
Syst., vol. 80, pp. 198–210, Mar. 2018. the M.Tech. degree (Hons.) from Guru Gob-
[52] H. S. Yin and R. Vatrapu, ‘‘A first estimation of the proportion of ind Singh Indraprastha University, Delhi, India,
cybercriminal entities in the bitcoin ecosystem using supervised machine in 2009, and the Ph.D. degree in wireless sensor
learning,’’ in Proc. IEEE Int. Conf. Big Data (Big Data), Dec. 2017,
network, in 2016. He is currently an Associate Pro-
pp. 3690–3699.
fessor with the Computer Science and Engineer-
[53] C. Qiu, F. R. Yu, F. Xu, H. Yao, and C. Zhao, ‘‘Blockchain-based dis-
ing Department, Institute of Technology, Nirma
tributed software-defined vehicular networks via deep Q-learning,’’ in
Proc. 8th ACM Symp. Design Anal. Intell. Veh. Netw. Appl. (DIVANet), University, Ahmedabad, Gujarat, India. He is also
New York, NY, USA, 2018, pp. 8–14. a Visiting Professor with Jan Wyzykowski Uni-
[54] J. Bhatia, R. Dave, H. Bhayani, S. Tanwar, and A. Nayyar, ‘‘Sdn-based versity, Polkowice, Poland, and also with the University of Pitesti in
real-time urban traffic analysis in vanet environment,’’ Comput. Commun., Pitesti, Romania. He has authored or coauthored more than 100 technical
vol. 149, pp. 162–175, Oct. 2019. research articles published in leading journals and conferences from the
[55] J. Bhatia, Y. Modi, S. Tanwar, and M. Bhavsar, ‘‘Software defined vehic- IEEE, Elsevier, Springer, and Wiley. Some of his research findings are
ular networks: A comprehensive review,’’ Int. J. Commun. Syst., vol. 32, published in top cited journals such as the IEEE TRANSACTIONS ON TVT,
no. 12, p. e4005, 2019. the IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, Applied Soft Computing,
[56] J. Vora, A. Nayyar, S. Tanwar, S. Tyagi, N. Kumar, M. S. Obai- the Journal of Network and Computer Application, Pervasive and Mobile
dat, and J. J. P. C. Rodrigues, ‘‘Bheem: A blockchain-based frame- Computing, the International Journal of Communication System, Telecom-
work for securing electronic health records,’’ in Proc. IEEE Globecom munication System, Computer and Electrical Engineering, and the IEEE
Workshops (GC Wkshps), Dec. 2018, pp. 1–6. SYSTEMS Journal. He has also published three edited/authored books with
[57] S. K. Singh, M. M. Salim, M. Cho, J. Cha, Y. Pan, and J. H. Park, ‘‘Smart International/National Publishers. He has guided many students leading to
contract-based pool hopping attack prevention for blockchain networks,’’ M.E./M.Tech. and guiding students leading to Ph.D. His current interest
Symmetry, vol. 11, no. 7, p. 941, 2019. includes wireless sensor networks, fog computing, smart grid, the IoT,
[58] W. M. Wang, H. Guo, Z. Li, Y. Shen, and A. V. Barenji, ‘‘Towards open and blockchain technology. He has been awarded best research paper
and automated customer service: A blockchain-based automl framework,’’ awards from IEEE GLOBECOM 2018, IEEE ICC 2019, and Springer
in Proc. 2nd Int. Conf. Comput. Sci. Appl. Eng. (CSAE), New York, NY,
ICRIC-2019.He was invited as a Program Chair, a Publications Chair, a Pub-
USA, 2018, pp. 27:1–27:6.
licity Chair, and a Session Chair in many International Conferences held in
[59] K. Singla, J. Bose, and S. Katariya, ‘‘Machine learning for secure device
North America, Europe, Asia, and Africa. He is an Associate Editor of IJCS,
personalization using blockchain,’’ in Proc. Int. Conf. Adv. Comput., Com-
mun. Informat. (ICACCI), Sep. 2018, pp. 67–73.
Wiley and Security and Privacy Journal, Wiley. He was invited as a Guest
[60] R. Gupta, S. Tanwar, S. Tyagi, N. Kumar, M. S. Obaidat, and B. Sadoun, Editors/Editorial Board Members of many International Journals, invited for
‘‘Habits: Blockchain-based telesurgery framework for healthcare 4.0,’’ keynote Speaker in many International Conferences held in Asia.
in Proc. Int. Conf. Comput., Inf. Telecommun. Syst. (CITS), Aug. 2019,
pp. 1–5.
[61] A. Kumari, S. Tanwar, S. Tyagi, N. Kumar, M. Maasberg, and QASIM BHATIA is currently pursuing the
K.-K. R. Choo, ‘‘Multimedia big data computing and Internet of Things degree with Nirma University, Ahmedabad, India.
applications: A taxonomy and process model,’’ J. Netw. Comput. Appl., His research interest includes machine learn-
vol. 124, pp. 169–195, 2018. ing, network security, fog computing, and cloud
[62] A. Mewada, S. Tanwar, and Z. Narmawala, ‘‘Comparison and evaluation computing.
of real time reservation technologies in the intelligent public transport
system,’’ in Proc. 5th Int. Conf. Parallel, Distrib. Grid Comput. (PDGC),
Dec. 2018, pp. 800–805.
[63] A. Kumari, S. Tanwar, S. Tyagi, and N. Kumar, ‘‘Fog computing for
healthcare 4.0 environment: Opportunities and challenges,’’ Comput. Elect.
Eng., vol. 72, pp. 1–13, Nov. 2018.
[64] S. Tanwar, K. Parekh, and R. Evans, ‘‘Blockchain-based electronic health-
care record system for healthcare 4.0 applications,’’ J. Inf. Secur. Appl.,
PRUTHVI PATEL is currently pursuing the degree
vol. 50, 2020, Art. no. 102407. with Nirma University, Ahmedabad, India. His
[65] Iota. Accessed: Nov. 4, 2019. [Online]. Available: https://www.iota.org/ research interest includes big data analytics, fog
[66] S. K. Singh, S. Rathore, and J. H. Park, ‘‘Blockiotintelligence: computing, and cloud computing.
A blockchain-enabled intelligent IoT architecture with artificial
intelligence,’’ Future Gener. Comput. Syst., to be published, doi:
10.1016/j.future.2019.09.002.
[67] K. Salah, M. H. U. Rehman, N. Nizamuddin, and A. Al-Fuqaha,
‘‘Blockchain for ai: Review and open research challenges,’’ IEEE Access,
vol. 7, pp. 10127–10149, 2019.

VOLUME 8, 2020 487

 S. Tanwar et al.: Machine Learning Adoption in Blockchain-Based Smart Applications

APARNA KUMARI received the M.Tech. degree WEI-CHIANG HONG (M’04–SM’10) is

from Jawaharlal Nehru University, Delhi, India, currently a Professor with the Department of
in 2012. She is currently pursuing the Ph.D. degree Information Management, Oriental Institute of
from the Department of Computer Science and Technology, Taiwan. His research interests mainly
Engineering, Institute of Technology, Nirma Uni- include computational intelligence (neural net-
versity, Ahmedabad, Gujarat, India. Her research works and evolutionary computation), and appli-
interest includes big data analytics, smart grid, cation of forecasting technology (ARIMA, support
blockchain technology, and cloud computing. vector regression, and chaos theory), and machine
learning algorithms. He serves as the program
committee for various international conferences
including premium ones such as IEEE CEC, IEEE CIS, IEEE ICNSC,
IEEE SMC, IEEE CASE, and IEEE SMCia. In May 2012, his article
had been evaluated as Top Cited Article 2007-2011 by Elsevier Publisher
PRADEEP KUMAR SINGH received the (Netherlands). In September 2012, once again, his article had been indexed
M.Tech. degree (Hons.) in CSE from GGSIPU, in ISI Essential Science Indicator database as Highly Cited Articles, in the
New Delhi, India, and the Ph.D. degree in com- meanwhile, he also had been awarded as the Model Teacher Award by Taiwan
puter science and engineering from Gautam Bud- Private Education Association. He is indexed in the list of Who’s Who in the
dha University (State Government University), World (25th–30th Editions), Who’s Who in Asia (2nd Edition), and Who’s
Greater Noida, India. He is currently working as Who in Science and Engineering (10th and 11th Editions). He has Google
an Assistant Professor (Senior Grade) with the scholar citations 5424, H-index 38, and i-10 Index 64 in his account. He is
Department of CSE, Jaypee University of Infor- a Senior Member of IIE. He is currently appointed as the Editor-in-Chief of
mation Technology (JUIT), Waknaghat. He has the International Journal of Applied Evolutionary Computation, in addition,
published nearly 85 research articles in various he serves as a Guest Editor for the Energies, and is appointed as an Associate
International Journals and Conferences of repute. He has received three Editor of the Neurocomputing, Forecasting, and the International Journal of
sponsored research projects grant from Governement of India and Gov- System Dynamics Applications.
ernment of HP worth Rs 25 Lakhs. He has edited total eight books from
Springer and Elsevier and also edited several special issues for SCI and SCIE
Journals from Elsevier and IGI Global. He has Google scholar citations 401,
H-index 12 and i-10 Index 15 in his account. He is having life membership
of Computer Society of India (CSI), a Life Member of IEI and promoted
to Senior Member Grade from CSI and ACM. He is an Associate Editor of
the International Journal of Information Security and Cybercrime (IJISC) a
scientific peer reviewed journal from Romania.

488 VOLUME 8, 2020