IOP Conference Series: Materials

Science and Engineering

PAPER • OPEN ACCESS You may also like

- Review—Power Sources for the Internet of
AIOT-Arch: Furthering Artificial Intelligence in Big Things
Abhi Raj and Dan Steingart
Data IoT Applications - An efficient hardware implementation of
SHA-3 using 3D cellular automata for
secure blockchain-based IoT systems
To cite this article: Sandra Said et al 2021 IOP Conf. Ser.: Mater. Sci. Eng. 1051 012008 Zied Guitouni, Noeman Ammar and
Mohsen Machhout

- Advanced error correction method for

quantum key distribution in IoT systems
Zied Guitouni, Sirine Maize, Mounir Zrigui
View the article online for updates and enhancements. et al.

This content was downloaded from IP address 118.70.128.236 on 29/10/2024 at 05:28

ICATAS-MJJIC 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1051 (2021) 012008 doi:10.1088/1757-899X/1051/1/012008

AIOT-Arch: Furthering Artificial Intelligence in Big Data

IoT Applications

Sandra Said.1*, Rowan Ihab.1, Ossama Hesham.1, Islam Abou Tabl.1, Nu

Maged.1, Sherin Youssef.1, Mazen Elagamy.1

1
Department of Computer Engineering, College of Engineering and technology Arab
Academy for Science and Technology, Alexandria, Egypt.

*Corresponding author: [email protected]

Abstract. Applications for the Internet of Things (IoT) have evolved in excessive numbers,
producing a vast amount of data needed for intelligent processing. The varying IoT
infrastructures such as cloud and IoT application layer protocol limitations in the
transmission/receiving of messages become the barriers in the implementation of intelligent IoT
apps. In this paper, we review the importance of Big data, cloud computing and fog computing
in IoT and the challenges of using machine learning in IoT. Finally, we discuss the general
statistics of using artificial intelligence in IoT applications.

1. Introduction:
The word "smart" fascinates us but the tools that we are using today are far from being smart like a
human being. Artificial intelligence is designed to make computers do human reasoning that why we
need machine learning [1] and data analysis [2] to be combined in one system. Machine learning creates
methods to make the network's component automatic and self-sufficient while data analysis analyzes
the data to identify the historical trends and be more effective and accurate in the future. IoT [3] is the
main object of this trend that will provide a word full of intelligent devices called “smart objects” [4]
integrated through the internet, Bluetooth or infrared. In this paper, we research and evaluate the role
of machine learning in promoting data analytics for the IoT systems.
2. Big Data in IoT environments:
Intrinsically, IoT data is a type of big data. The widespread use of sensors for data collection, the
utilization of the data collected over long periods of time and the need to evaluate them on a scale to
help decision taking means that they overlap with many dimensions of big data [5]. In the following
section we will describe the main characteristics of big data found in IoT and how these characteristics
are considered as a challenge when using traditional machine learning techniques. Figure 1 shows the
IoT data characteristics, the left part of the figure concerning big data characteristics will be discussed
in the following section.

3. Characteristics of big Data which lead to inefficiency of Traditional Machine Learning

Techniques:
3.1. Velocity

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
ICATAS-MJJIC 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1051 (2021) 012008 doi:10.1088/1757-899X/1051/1/012008

velocity is one of the main properties of big data hence IoT data. The velocity dimension refers to how
quickly the data is being produced and how fast the application needs to process it. "Fast data" is
endemic to IoT, since the generated data is in most cases a continuous and unbounded stream from a
multitude of sensors [6]. The velocity of IoT data is affected by many factors. It’s affected directly by
the sampling rate and the number of sensors in a given environment. Very high sampling rates can also
often lead to redundancy if the observations are changing slowly. Nonetheless, a certain amount of
redundancy will be optimal with some degree of trust for detecting the events with some level of
confidence [5].
3.2. Volume
volume of data is not less important than velocity. Big data implies enormous volumes of data. One
view is the volume of data generated per sensor and the number of sensors (distributed). But these
sensors are constantly observing, and the streaming data is potentially unbounded.

3.3. Variety
The variety component of big data is representative of the various potential types of data representations
and protocols. IoT data heterogeneity is common, and can be due to different data sources, data types,
or networks. IoT applications collect data from a wide range of sensors and in a wide range of formats
due to the variety of environment variables being monitored.

3.4. Veracity
veracity refers to the biases, noise and abnormality in data. The question here is whether the data that
is being stored, and mined is relevant to the problem being analyzed. Figure 2 shows the characteristics
of Big Data [31].

Figure 1. IoT data characteristics Figure 2. Characteristics of big data

4. Why Cloud Computing and Fog Computing?

4.1. Role of Cloud Computing in IoT:
Cloud computing is the on-demand of computer resources like data storage and computing power over
the internet. Companies prefer to use a cloud service provider [7] rather than using their own data
centers. Cloud computing helps in the storage and the analysis of the large amount of data generated by
the IoT applications and connects objects and people together to provide a high visibility [8]. IoT Links
existing devices and smart devices to the internet and reduces gaps between Information Technology
and Operational technology teams to create a single systems/data view. Nowadays the deployment of
comprehensive hardware and the management of networks in IoT systems is not the main problem with
the help of cloud computing.

4.2. Role of Fog Computing in IoT:

Fog computing is a platform that uses edge devices to conduct a large amount of local and routed
storage, computation and communication over the internet [7]. It is a distributed environment that is

2
ICATAS-MJJIC 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1051 (2021) 012008 doi:10.1088/1757-899X/1051/1/012008

related to the cloud computing and IoT. Figure 3 shows different IoT components, and how analytics
apply to these components [9]. Fog computing aims to form low latency network links between devices
and the endpoints of their analyses and reduces the bandwidth required [10]. An additional advantage
is the advanced security features that users can implement in a fog network such as network traffic
segmentation and virtual extension of firewalls to provide network security.

5. IoT-Generated Data Processing for Machine Learning Analysis

Because IoT devices send data repetitively, big data challenges are experienced [14,15]. To get past this
problem, distributed processing can be leveraged to partition data into chunks and assign each partition
to a different computer for processing. Different frameworks can be used to achieve distributed
processing like Hadoop and Spark, which can be used in machine learning. When using distributed
computing, the following benefits are gained:
1) Network load is lessened.
2) Faster processing of data.
3) Lower energy consumption.
4) Processing of huge volumes of data becomes feasible. [5]

6. Challenges in Developing Intelligent IoT Solutions

6.1. Security
Communication technologies used by IoT devices such as Zigbee, Z-Wave and Wi-Fi have their own
limitations in terms of security, which hinders the security of IoT applications [11]. The widespread use
of IoT devices increases the vulnerability of an IoT system to attacks. Moreover, devices used in IoT
systems usually are resource limited, which makes the use of advanced security techniques to protect
IoT devices hardly achievable.

6.2. Performance
Some machine learning algorithms like Deep Convolutional Neural Networks can achieve high
accuracies, but they have high computational and memory requirements, which make it hard to
implement and sometimes infeasible on the resource-constrained devices used in IoT systems, for
example in safety-critical applications like autonomous driving, which requires real-time image
processing [11]. The following table 1 shows frame rates of YOLO algorithm, which is a state-of-the-
art algorithm used in object detection applications, accomplished using Nvidia Jetson TX1 embedded
module [12], which is made to be used in visual computing applications [35].

Algorithm Frame Rate

Fast YOLO 17.85

O-YOLOv2 11.8

YOLOv2 5.4
Table 1. Results of testing YOLO algorithms on Nvidia Jetson TX1
As shown in the Table 1, the frame rate didn’t exceed 17.85, which is too slow for applications that
require real-time processing, which requires a frame rate of 30 [11].

6.3. Reliability
This is another core issue in machine learning algorithms. Machine learning algorithms like neural
networks are not entirely accurate, which makes it unreliable for applications with high sensitivity to
accuracy, like self-driving vehicles, and cancer diagnosis. Moreover, deep learning is being introduced
in many industrial applications including data mining & analytics, where certain industrial standards
need to be met, like IEC 61508 [13], which specifies reliability specifications to be met in industrial
applications. Therefore, deep learning-based systems reliability needs to be ensured [13].

3
ICATAS-MJJIC 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1051 (2021) 012008 doi:10.1088/1757-899X/1051/1/012008

7. Deep Learning Machine Learning for IOT:

Nowadays, the trend of Artificial intelligence has opened new windows to build a variety of deep
learning frameworks. Deep learning platforms e.g. TensorFlow are frequently used for text analysis in
multimedia systems. The texts are analyzed using deep learning techniques such as CNN and RNN
architectures. Some examples include detection of actions in video, recognition of disasters in the web
crawled datasets. Figure 4 shows the difference between machine learning and deep learning.

Figure 3. Convergence of IoT architecture Figure 4. The difference between Machine

and data analytics: an illustration learning and Deep learning

7.1. Deep learning Layered Architecture:

The deep learning architecture is an 8 layers convolutional neural network (CNN) composed of 3
convolutional layers, three max-pooling (subsampling) layers and two fully connected layers as shown
in Figure 5.

Figure 5. Deep learning architecture

7.1.1. Convolutional Layer

The major building blocks used in convolutional neural networks are the conventional layers. A
convolution is simply applying a filter to an input resulting in an activation. Repeated application of the
same filter to an input results in an activation map called a feature map, which indicates the location
and strength of a detected feature in an input such as an image[16].The innovation of convolutional
neural networks is the ability to automatically learn, under the constraints of a specific predictive
modelling problem, such as image classification, a large number of filters in parallel specific to a
training dataset. The result is highly specific features, which can be detected on input images anywhere
[17].

7.1.2. Pooling or Subsampling layer

Convolution layer in CNN is often immediately followed by the pooling or subsampling layer. Its role
is to down sample the output of a convolution layer along both the height and width dimensions of the

4
ICATAS-MJJIC 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1051 (2021) 012008 doi:10.1088/1757-899X/1051/1/012008

space [19]. For example, an operation of 2 x 2 pooling on top of 12 feature maps will produce an output
size tensor [16 x 16 x 12].

7.1.3. Fully Connected Layer

Fully Connected Layer can be defined as a feed forward neural network. Fully Connected Layers are
considered from the last few layers in the network [18]. The output from the final pooling or
Convolutional layer is the input to the fully connected layer, which is flattened first then fed into this
last layer [19].

8. Comparisons between different machine learning Techniques for IoT Application:

Use Case for IoT Machine learning Advantages Disadvantages

algorithm
(Data processing
tasks)

Traffic and Air K-mean Better for huge K- value difficult

Control monitoring (Clustering) variables and also and difficult for
produces tighter data of different
clusters densities

Agriculture data Naive Bayes Simple to Assumption can be

analysis (Classification) [21] implement. Needs wrong
less training data

Lameness detection K-Nearest Neighbor It will be helpful The value for k

(Classification) [23] even if the training must be known as
data is noisy accurately

System to estimate Support Vector Data can be Selection of a good

personal thermal Machine visualized in more kernel value is not
comfort (Classification) [20] than 2 dimensions. easy, and also, if
Decision taking the data set is large
system is nearly then it will take
perfect more time for
training

Monitoring Public Principal Lack of The covariance

Places, Useful in Component Analysis redundancy, matrix is difficult to
Fault Detection. (Feature extraction) Reduced evaluate and even
complexity in the simplest
images invariance could
not be captured

Low energy Neural Network Parameters are hard

consumption, (Regression/Classifi to interpret.
Neural Network
Healthcare Analysis cation/ It becomes very
and Forecasting can model
complex
sometimes.

5
ICATAS-MJJIC 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1051 (2021) 012008 doi:10.1088/1757-899X/1051/1/012008

Clustering/Feature nonlinear and Needs a large

extraction) complex relations. dataset

Predicting energy Random Forrest considered as good Noise in the

consumption using (Classification/Regre for takings datasets
a smart-metering ssion) [24] thousands of input
architecture variables

Predict the energy Linear Regression simple to outliers can have

usage (Regression) [22] implement. huge effects.
susceptible to over- assumes a linear
fitting but it can be relationship
avoided using some
dimensionality
reduction
techniques

Temperature and Support Vector Works with non- Not a familiar

humidity data Regression linear problems. mode.
prediction (Regression) Not biased by Quite hard to
outliers understand

customer attrition Bagging customer attrition final prediction

prediction (Classification/Regre prediction. based on the mean
ssion) [25] reduces model predictions.
over-fitting.
missing values in
the dataset do not
affect the
performance

Table 2. Comparison between different machine learning techniques in IoT applications

9. Application Case-Study in Smart Cities:

9.1. Smart City
Smart cities use IoT devices such as sensors to collect data and cloud computing systems to store and
analyze data [26]. The IoT devices are connected to each other using Wi-Fi connection or Bluetooth or
infrared to create a huge single network that improves the citizen's lifestyle and public services such as
energy consumption and water management. Smart city uses the data collected from IoT devices to
monitor and manage the city's assets. Smart City is the main use case of IoT applications because of the
following three reasons: Firstly, 60% of the publications examined is on the Smart City sector, secondly,
smart City includes many other IoT use cases and finally, huge number of open datasets for Smart City
applications exists that are easily accessible to researchers.

9.2. Case-Study of IoT in smart cities

9.2.1. Smart Road
Municipalities use IoT systems to provide smart road that helps people to navigate through places
rapidly and safely. Different Sensors and Data collected from smart phone's GPS help smart road
applications to determine the vehicles number, position and speed. Smart road manages the traffic light

6
ICATAS-MJJIC 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1051 (2021) 012008 doi:10.1088/1757-899X/1051/1/012008

that is connected to a cloud platform depending on the actual traffic situation [27]. By the time, the
smart road system can predict the time and the location of the traffic and prevent that automatically.

9.2.2. Smart parking

Smart parking applications create a real time map of parking spots using sensors implemented in the
parking spots and data collected from driver's GPS [28]. This data is sent to the server which is a cloud
system that collects and analyze the data. After analysis, the server sends notifications of the available
parking spots to the driver’s smart phone application. The drivers are capable of paying by cash or visa
(using the smart phone's application) as shown in Figure 6.
9.2.3. Smart Street lighting
Smart Street lighting system consists of sensors in the lighting poles. The poles are connected together
using Wi-Fi connection. The data collected from these sensors is sent to the cloud system (control and
management system) that analyzes the data and control the street lighting schedule [29] as shown in
Figure 7. The sensors collect data concerning the movement of citizens in the road, the movement of
cars and the degree of illuminance. The lighting poles gets brighter when a walker crosses the road or
a bus arrives to the bus stop.

Figure 6. Smart Parking System Overview Figure 7. Street lightening System overview

10. General statistics:

10.1. Datasets:

10.1.1. Transportation Mode Detection

The dataset of the transportation mode is collected using smartphones. Three main sensors are used to
collect data which are the accelerometer, gyroscope and sound. The idea is to detect/recognize what
type of vehicle the user uses, as there are five classes which are car, bus, train, walking and still. The
process is considered as (HAR) Human Activity Recognition (HAR) [30]. The Specifications of this
dataset are shown in Table 3.

Number of Samples Number of Features Number of Target

Classes

5893 14 5
Table 3. Specifications of the Transportation Mode detection dataset
10.1.2. Rain Prediction
The dataset is collected from Australia through the observation of weather stations. The dataset includes
many attributes such as: rainfall, evaporation, sunshine, temperature, direction, wind speed, pressure,

7
ICATAS-MJJIC 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1051 (2021) 012008 doi:10.1088/1757-899X/1051/1/012008

clouds and humidity. The dataset is used to predict whether there will be rain the next day or not [31].
The Specifications of this dataset are shown in Table 4.

Number of Samples Number of Features Number of Target

Classes

142000 24 2
Table 4. Specifications of the Rain detection dataset

10.2. Performance Evaluation Metrics:

Precision 𝑇𝑃 number of relevant selected data items

𝑇𝑃 + 𝐹𝑃 that are relevant.

Recall 𝑇𝑃 number of relevant data items that are

𝑇𝑃 + 𝐹𝑁 selected.

F1-score 𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 ∗ 𝑅𝑒𝑐𝑎𝑙𝑙 The harmonic mean of precision and

2∗
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 + 𝑅𝑒𝑐𝑎𝑙𝑙 recall.

Accuracy 𝑇𝑃 + 𝑇𝑁 The ratio of accurately classified data

𝑇𝑃 + 𝐹𝑁 + 𝑇𝑁 + 𝐹𝑃 items to the total number of observations.

Confusion Allows to measure Recall, Precision, Accuracy, and AUC-ROC

Matrix curve. Its main usage is when a classification problem has two or
more outputs.

ROC-AUC Used for discriminating the positive and negative classes in binary
score classification and demonstrates how good a model is.
Table 5. Performance Metrics
10.3. Results
10.3.1. Performance comparison of the algorithms on Transportation Mode Detection dataset:
The accuracy rate is the most reliable measure in this case because for each target class in the dataset
there are equal numbers of labels which means that the dataset is balanced [32] as shown in Table 6.

Algorith Precision Recall F1-Score Training Test Set Executio

ms % % % Set Accuracy n Time
Accurac (Avg/Highe
y st) %
%

LR 63 62 62 64 63/65 0.68

KNN 83 83 83 87 80/82 0.005

DT 80 79 79 100 76/79 0.067

RF 87 87 87 100 85/87 0.71

SVM 80 80 80 88 79/81 9.9

CNN 82 80 80 83 82/82 16.03

Table 6. Performance Metrics of different algorithms on Transportation Mode Detection dataset

8
ICATAS-MJJIC 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1051 (2021) 012008 doi:10.1088/1757-899X/1051/1/012008

10.3.2. Performance comparison of the algorithms on Rain Prediction dataset:

The Rain Prediction dataset has 24 features, three of the features are categorical features each one of
them has 16 different values, for that kind of information the dataset required much more pre-processing
tasks. One-hot encoding techniques are applied to the dataset to convert the categorical features to
numeric ones, which lead to tripling the number of features [33] as shown in Table 7.

Algorith Precision Recall F1-Score Training Test Set Executio

ms % % % Set Accuracy n Time
Accuracy (Avg/Hig
% hest)
%

LR 72 79 74 85 85/85 8.2

KNN 76 72 74 90 81/83 10.9

DT 70 71 70 100 79/80 2.22

RF 80 71 74 99 85/85 1.42

ANN 82 74 77 86 85/86 35.00

CNN 78 73 75 85 84/85 24.8

Table 7. Performance Metrics of different algorithms on Rain Detection dataset

11. Conclusion:
Nowadays, the IoT applications are integrated in everything. This paper shows the importance of the
usage of artificial intelligence in Big data IoT applications using our proposed framework and shows
the power of machine learning in providing benefits to the consumers. The proposed framework is
examined on two datasets (Transportation Mode Detection dataset and Rain Prediction). When using
different machine learning techniques with the first dataset, it gives best accuracy of 85/87% with RF
algorithm but when using these same techniques with the second dataset, it gives best accuracy of
85/86% with the ANN algorithm. Our innovative architecture is opening up new opportunities for
potential machine-to- machine communications work.

References
[1] R. S. Michalski, J. G. Carbonell, and T. M. Mitchell, Machine Learning: An Artificial
Intelligence Approach. Springer Science & Business Media, 2013.
[2] I. H. Witten and E. Frank, Data Mining: Practical Machine Learning Tools and Techniques.
Morgan Kaufmann, 2016.
[3] Q. F. Hassan, A. R. Khan, and S. A. Madani, Internet of Things: Challenges, Advances, and
Applications. Chapman & Hall/CRC Computer and Information Science Series, CRC Press,
2017.
[4] G. Fortino and P. Trunfio, Internet of Things based on Smart Objects: Technology, Middleware
and Applications. Springer, 2014.
[5] Mohammad Saeid Mahdavinejad, Mohammadreza Rezvan, Mohammadamin Barekatain,
Peyman Adibi, Payam Barnaghi, Amit P. Sheth; Machine Learning for Internet of Things
Data Analysis: A Survey. 2018
[6] Nashez Zubair;Niranjan A;Kiran Hebbar;Yogesh Simmhan: Characterizing IoT Data and its
Quality for Use. 2019
[7] Bonomi F, Milito R, Zhu J, Addepalli S (2012) Fog computing and its role in the internet of
things. In: Proceedings of the First Edition of the MCC Workshop on Mobile Cloud
Computing, ACM, New York, NY, USA, MCC ’12, pp 13–16
[8] Joshi J, Reddy J, Reddy P, Agarwal A, Agarwal R, Bagga A, Bhargava A (2016) Cloud

9
ICATAS-MJJIC 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1051 (2021) 012008 doi:10.1088/1757-899X/1051/1/012008

computing based smart garbage monitoring system. In: 2016 3rd international conference on
electronic design (ICED), IEEE, pp 70–75
[9] Aazam M, Zeadally S, Harras KA (2018) Offloading in fogcomputing for iot: review, enabling
technologies, and research opportunities. Future Generat Comput Syst 87:278–289
[10] Fog computing and the internet of things: extend the cloud to where the things are (2015).
https://www.cisco.com/c/dam/en_us/solutions/trends/iot/docs/computing-overview.pdf.
Accessed 18 Oct 2019
[11] Fatima Hussain, Rasheed Hussain, Syed Ali Hassan, and Ekram Hossain, Machine Learning in
IoT Security: Current Solutions and Future Challenges, IEEE 2020
[12] Shafiee, Mohammad Javad, et al. "Fast YOLO: A Fast You Only Look Once System for Real
time Embedded Object Detection in Video." arXiv preprint arXiv:1709.05943 (2017).
[13] IEC 61508 2016. Functional Safety and IEC 61508. (2016). Retrieved Oct. 2016 from
http://www.iec.ch/functionalsafety/
[14] A. Joakar, A Methodology for Solving Problems with DataScience for Internet of Things, Open
Gardens(blog)(July21,2016),http://www.opengardensblog.futuretext.com/archives/2016/07/
a-methodology-for-solvingproblems-with-datascience-for-internet-of-things.html
[15] H. Zhao and C. Huang, A data processing algorithm in epc internet of things, Cyber-enabled
Distributed Computing and Knowledge Discovery (CyberC), 2014 International Conference
on, IEEE (2014), pp. 128-131.
[16] E. Shelhamer, J. Long, and T. Darrell. Fully convolutional networks for semantic segmentation.
IEEE Trans Pattern Anal Mach Intell, 39(4):640–651, 2017.
[17] Farabet, C.; Couprie, C.; Najman, L.; LeCun, Y. Learning hierarchical features for scene labeling.
IEEE Trans. Pattern Anal. Mach. Intell. 2013, 35, 1915–1929. [CrossRef] [PubMed]
[18] Simonyan, K.; Zisserman, A. Very deep convolutional networks for large-scale image
recognition. arXiv 2014, arXiv:1409.1556.
[19] Krizhevsky, A.; Sutskever, I.; Hinton, G.E. Imagenet classification with deep convolutional
neural networks. In Proceedings of the 25th International Conference on Neural Information
Processing Systems, Lake Tahoe, NV, USA, 3–6 December 2012; pp. 1097–1105.
[20] N. Cristianini, J. Shawe-Taylor, An introduction to support vector machines and other kernel-
based learning methods, Cambridge university press, 2000.
[21] McCallum, K. Nigam, et al., A comparison of event models for naïve bayes text classi_cation,
in: AAAI-98 workshop on learning for text categorization, Vol. 752, Citeseer, 1998, pp. 41-
48.
[22] Neter, M. H. Kutner, C. J. Nachtsheim, W. Wasserman, Applied linear statistical models, Vol. 4,
Irwin Chicago, 1996.
[23] T. Cover, P. Hart, Nearest neighbor pattern classi_cation, IEEE transactions on information
theory 13 (1) (1967) 21-27.
[24] Breiman, Random forests, Machine learning 45 (1) (2001) 5-32.
[25] Breiman, Bagging predictors, Machine learning 24 (2) (1996) 123-140.
[26] R. Petrolo, V. Loscri, N. Mitton, Towards a smart city based on cloud of things, a survey on the
smart city vision and paradigms, Transactions on Emerging Telecommunications
Technologies.
[27] M. Vanis and K. Urbaniec, “Employing Bayesian Networks and conditional probability functions
for determining dependences in road traffic accidents data,” 2017 Smart City Symposium
Prague (SCSP), May 2017.
[28] Prabhu Ramaswamy; IoT smart parking system for reducing green house gas emission; 2016
International Conference on Recent Trends in Information Technology (ICRTIT)
[29] N. Yoshiura, Y. Fujii, and N. Ohta. Smart street light system looking like usual street lights based
on sensor networks. In International Symposium on Communications and Information
Technologies (ISCIT), pages 633–637, Sept 2013
[30] D. Kleyko, R. Hostettler, W. Birk, and E. Osipov, "Comparison of machine learning techniques
for vehicle classification using road side sensors," in Proceedings of the IEEE 18th
International Conference on Intelligent Transportation Systems, 2015, pp. 572-577.
[31] Y. K. Ever, K. Dimililer, and B. Sekeroglu, "Comparison of Machine Learning Techniques for

10
ICATAS-MJJIC 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1051 (2021) 012008 doi:10.1088/1757-899X/1051/1/012008

Prediction Problems," in Workshops of the International Conference on Advanced

Information Networking and Applications, 2019, pp. 713-723.
[32] F. Alam, R. Mehmood, I. Katib, and A. Albeshri, "Analysis of eight data mining algorithms for
smarter Internet of Things (IoT)," Procedia Computer Science, vol. 98, 2016, pp. 437-442.
[33] C. Carpineti, V. Lomonaco, L. Bedogni, M. Di Felice, and L. Bononi, "Custom Dual
Transportation Mode Detection by Smartphone Devices Exploiting Sensor Diversity," in
Proceedings of the IEEE International Conference on Pervasive Computing and
Communications Workshops (PerCom Workshops), 2018, pp. 367-372.
[34] "Innovations in Smart Cities Applications Edition 2", Springer Science and Business Media LLC,
2019.
[35] Jetson TX1 Module, Nvidia Developer, https://developer.nvidia.com/embedded/jetson-tx1

11