IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 14, No. 2, April 2025, pp. 1401~1409

ISSN: 2252-8938, DOI: 10.11591/ijai.v14.i2.pp1401-1409  1401

A comparative analysis of exponential smoothing method and

deep learning models for bitcoin price prediction

Nrusingha Tripathy1, Debahuti Mishra1, Sarbeswara Hota2, Mandakini Priyadarshani Behera3,

Gobinda Chandra Das4, Sasanka Sekhar Dalai1, Subrat Kumar Nayak1
1
Department of Computer Science and Engineering, Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar, India
2
Department of Computer Application, Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar, India
3
School of Computer Engineering, Kalinga Institute of Industrial Technology (Deemed to be University), Bhubaneswar, India
4
Department of Computer Science and Applications, KL Deemed to be University, Guntur, India

Article Info ABSTRACT

Article history: Blockchain technology is the foundation of cryptocurrencies, which are
virtual currencies. The decentralized nature of cryptocurrencies has resulted
Received Mar 15, 2024 in a significant reduction of central authority over them, which has
Revised Oct 23, 2024 implications for global trade and relations. The need for an effective model
Accepted Nov 14, 2024 to anticipate the price of cryptocurrencies is essential due to their wide
variations in value. Due to the shortcomings of conventional production
forecasting, in this work, four distinct models were used. The deep learning
Keywords: models are the long short-term memory (LSTM) and bidirectional long
short-term memory (Bi-LSTM), and both the Facebook-Prophet and
Bi-LSTM Silverkite support the exponential smoothing technique. Silverkite is
Cryptocurrency designed to handle a wide range of time series forecasting tasks. Considering
Facebook-Prophet past bitcoin information from January 2012 to March 2021, a period of nine
Financial data analysis years, we looked at the models. The Bi-LSTM model yields a 7.073 mean
Long short-term memory absolute error (MAE) and a 3.639 root mean squared error (RMSE). The
Silverkite Bi-LSTM model identifies the deviations that might draw attention and avert
any problems.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Nrusingha Tripathy
Department of Computer Science and Engineering, Siksha ‘O’ Anusandhan (Deemed to be University)
Bhubaneswar 751030, India
Email: [email protected]

1. INTRODUCTION
The decentralized network of computers that power cryptocurrencies collaborate to validate the log
transactions. Therefore, the money is no longer subject to the rheostat of authority like a bank or the
government. Cryptocurrency transactions are documented on a blockchain, an immutable, decentralized
ledger [1], [2]. To enable private and secure transactions, public and private keys are employed.
Lahmiri and Bekiros [3] testing the nonlinearity, produced data showing fractal dynamics, extended memory,
and self-similarity in all digital currency time series. There is a cap on the total quantity of coins or tokens
that may ever be issued for several cryptocurrencies. As an illustration, the maximum quantity of Bitcoin
coins is set at 21 million, a measure intended to produce scarcity and perhaps boost value over time.
Transactions with cryptocurrencies are often faster and may be less expensive than those using traditional
banking systems, and it can be sent and received anywhere on the globe.
Cryptocurrencies are a popular investment option for investors. In addition to being used for online
payments and remittances, digital assets are also utilized for other practices. According to Sun et al. [4], the
prediction's performance is impacted by the cryptocurrency's overall strength. This can reduce risks and

Journal homepage: http://ijai.iaescore.com

1402  ISSN: 2252-8938

provide investors with excellent guidance when building a bitcoin portfolio. Nonetheless, the market for
cryptocurrencies is renowned for its unpredictability, difficulties with regulations, and constant technical
advancements [5]. Predictions may be used by traders and investors to create hedging strategies, which try to
reduce total risk by offsetting possible losses in one area of the portfolio with profits in another. Predictions
about cryptocurrencies might be useful to traders when deciding whether to enter or leave the market. This
might be essential for optimizing gains or reducing losses under quickly shifting market circumstances [6]‒[8].
In this work, long short-term memory (LSTM) and bidirectional-long short-term memory
(Bi-LSTM) are employed that automatically extract pertinent characteristics. This is helpful because it
enables the model to derive accurate depictions from the input data, especially when working with big and
complicated datasets [9]. Similarly, the exponential smoothing methods Facebook-Prophet (FB-Prophet) and
Silverkite are castoff; these models are straightforward and obvious methods that give historical observations
exponentially diminishing weights.

2. RESEARCH METHODOLOGY
We are using the real time update provided by the Bitcoin exchange dataset. The dataset consists of
the following eight parameters: timestamp, open, high, low, close, volume in currency, volume in bitcoin,
and weighted price. The correlation heatmap of Bitcoin dataset is given in Figure 1. The timestamp data
columns include no transactions or activity are occupied with not a number (NaNs). If a timestamp is absent,
it may be due to an unanticipated technical fault in data reporting or collection [10]. From Kaggle contests,
we amassed bitcoin historical data. Here, the historical bitcoin data is at 1-minute interludes.

Figure 1. Correlation heatmap of bitcoin data

2.1. Long short-term memory

Prices of cryptocurrencies frequently show long-term patterns and interdependence. In order to
overcome the vanishing gradient issue that plagues conventional recurrent neural networks (RNNs), LSTMs
are made to be able to learn and retain data across longer sequences [11]. LSTMs do not require human
feature engineering since they can automatically extract relevant characteristics from the input data [12]. For
bitcoin prediction jobs involving complicated and high-dimensional data, this is advantageous. In (1) to (6)
shows the simple LSTM formulations. Since LSTMs can tolerate varying intervals of time between
observations, they are a good fit for data on cryptocurrencies, since price fluctuations may not happen at
regular intervals. Where 𝑓𝑡 is the forget gate, 𝑖𝑡 is the input gate, 𝑜𝑡 is the output, 𝑐̃𝑡 is the new candidate
value, ℎ𝑡 is the hidden state, 𝜎 is the sigmoid instigation function, and tanh is the hyperbolic tangent
instigation function.

𝑓𝑡 = 𝜎(𝑊𝑓 . [ℎ𝑡−1 , 𝑥𝑡 ] + 𝑏𝑓 ) (1)

𝑖𝑡 = 𝜎(𝑊𝑖 . [ℎ𝑡−1 , 𝑥𝑡 ] + 𝑏𝑖 ) (2)

Int J Artif Intell, Vol. 14, No. 2, April 2025: 1401-1409

Int J Artif Intell ISSN: 2252-8938  1403

𝑜𝑡 = 𝜎(𝑊𝑜 . [ℎ𝑡−1 , 𝑥𝑡 ] + 𝑏𝑜 ) (3)

𝑐̃𝑡 = tanh (𝑊𝑐 . [ℎ𝑡−1 , 𝑥𝑡 ] + 𝑏𝑐 ) (4)

𝑐𝑡 =𝑓𝑡 . 𝑐𝑡−1 + 𝑖𝑡 . 𝑐̃𝑡 (5)

ℎ𝑡 = 𝑜𝑡 . tanh (𝑐𝑡 ) (6)

2.2. Facebook-Prophet
Prophet can handle holidays and seasonality on a weekly and annual basis. This helps to capture
recurrent trends and the possible influence of holidays or unique occasions on price movements in the context
of bitcoin prediction [13]. Potential changepoints, or locations where the time series shows a notable shift in
trend, are automatically detected by Prophet. In (7) to (9) shows the FB-Prophet formulations. This helps to
record abrupt fluctuations in the price of cryptocurrencies. Because Prophet's settings are simple to understand
and adjust, even individuals without a lot of experience with time series modelling may use it [14], [15].

2𝜋𝑖𝑡 2𝜋𝑖𝑡
𝑠(𝑡) = ∑𝑁
𝑖=1(𝑈𝑖 . sin ( ) + 𝑉𝑖 . cos ( )) (7)
𝑍 𝑍

Where 𝑠(𝑡) is the seasonality at time 𝑡, 𝑁 is the number of seasonal components 𝑈𝑖 and 𝑉𝑖 are the
coefficients for ith component, 𝑍 is the period of seasonality.

𝑔(𝑡) = 𝑘(𝑡) + ∑𝑁
𝑛−1(𝛿𝑛 . ℎ(𝑡 − 𝑇𝑛 )) (8)

Where 𝑔(𝑡) is the trend at time 𝑡, 𝑘(𝑡) is the baseline growth, 𝑁 is the numeral of change points, 𝛿𝑛 is the
magnitude of n-th change point. During model training, managing holidays entails defining holidays and
how they affect the time series data. Prophet's architecture allows for the inclusion of holidays and other
special occasions in the forecasting process. 𝑦(𝑡) is the observed value at time 𝑡, 𝑔(𝑡) is the leaning, 𝑠(𝑡) is
the seasonality, ℎ(𝑡) is the day off effect, and ∈𝑡 is the fault term. The formulation for handling holidays is
given in (9).

𝑦(𝑡) = 𝑔(𝑡) + 𝑠(𝑡) + ℎ(𝑡) +∈𝑡 (9)

2.3. Silverkite
LinkedIn releases the time-series prediction library Greykite to simplify forecasting for data
scientists. The primary forecasting algorithm in this package is Silverkite, an automated prediction technique.
LinkedIn developed Greykite to help employees make informed decisions using time-series forecasting
models. A multitude of features are automatically generated by Silverkite, capturing various properties of the
time sequence. Silverkite's incorporation of various periodic patterns, trends, and unique occurrences enables
flexible forecasting. This adaptability helps to capture the dynamic and diversified character of the bitcoin
markets. Because Silverkite is built to withstand outliers and missing data, it can produce accurate forecasts
even in situations where the supplied data is incomplete. In (10) and (11) shows the Silverkite forecast
formulation. By optimizing for forecast accuracy, Silverkite automatically chooses the optimal forecasting
model from a group of candidate models. This can be helpful in the prediction of cryptocurrencies, where
selecting a suitable forecasting model might be difficult. Where 𝑦̂𝑡 is the calculated value at time 𝑡, 𝑏0 is the
intercept, 𝑏𝑖 , 𝑏𝑗 , 𝑏𝑘 are coefficients, ∈𝑡 is the error term.

𝑦̂𝑡 = 𝑏0 + ∑𝑁 𝑁 𝑁
𝑖=1 𝑏𝑖 . 𝑠𝑖,𝑡 + ∑𝑗=1 𝑏𝑗 . ℎ𝑗,𝑡 +∑𝑘=1 𝑏𝑘 . 𝑐𝑘,𝑡 +∈𝑡 (10)

𝑚𝑠 =𝑎𝑚 (∑𝑇𝑡=1(𝑦𝑡 − 𝑦̂𝑚,𝑡 )2 ) (11)

2.4. Bidirectional-long short-term memory

The price fluctuations of cryptocurrencies frequently show long-term tendencies and relationships.
Long-term trends in the historical pricing data may be found and used with Bi-LSTM's capacity to record
knowledge gathered from current and future time steps [16]. Numerous variables impact cryptocurrency
markets, and there may be nonlinear interactions between these elements. Being a kind of neural network,
Bi-LSTM may adjust to the dynamic and frequently unpredictable character of cryptocurrency markets by
learning intricate nonlinear correlations in the data [17], [18]. In (12) to (18) shows the Bi-LSTM forecast
formulation.

A comparative analysis of exponential smoothing method and deep learning … (Nrusingha Tripathy)
1404  ISSN: 2252-8938

𝑓⃖𝑡 =𝜎(𝑅𝑓𝐿⃖ [ℎ𝑡−1

𝐿
, ℎ𝑡𝐿−1 ] + 𝑠𝑓𝐿⃖ ) (12)

⃖𝑖𝐿𝑡 =𝜎(𝑅𝑓𝐿⃖ [ℎ𝑡−1

𝐿
, ℎ𝑡𝐿−1 ] + 𝑠𝑓𝐿⃖ ) (13)

𝑜⃖𝑡𝐿 =𝜎(𝑅𝑜𝐿⃖ [ℎ𝑛−1

𝐿
, ℎ𝑡𝐿−1 ] + 𝑠𝑜𝐿⃖ ) (14)

𝑐⃖𝑡𝐿 =𝑓⃖𝑡𝐿 .𝑐⃖𝑛−1

𝐿
+ ⃖𝑖𝐿𝑡 . 𝑐⃖𝑛𝐿⃖ −1 (15)

𝑐⃖𝑡𝐿 =tanh (𝑅𝑐𝐿⃖ [ℎ𝑛−1

𝐿
, ℎ𝑡𝑀−1 ] + 𝑠𝑐𝐿⃖ ) (16)

⃖ 𝑡 = 𝑜⃖𝑡𝐿 . tanh (𝑐⃖𝑡𝐿 )

ℎ (17)

⃖ 𝑡, ℎ
𝑦𝑡 = 𝑅[ℎ ⃖ 𝑡] + 𝑠 (18)

3. PROPOSED MODEL
The dynamics of cryptocurrency marketplaces result in fluctuating values due to many variables.
As additional data becomes available, deep learning models may update their internal representations to
reflect the shifting market conditions [19]. Experts at automatically extracting pertinent characteristics from
unprocessed data are deep learning algorithms. In the domain of cryptocurrency forecasting, where intricate
correlations exist between attributes and price fluctuations, this is particularly advantageous [20], [21].
The proposed workflow diagram is given in Figure 2. Every data point in the collection has a time stamp or
period attached to it that indicates the exact moment the observation was made. The consequences of the
Dickey-Fuller test (DFT) are shown in Table 1.
Time series data may be studied to uncover underlying dynamics and forecast future values since it
frequently shows patterns, trends, or seasonality. Cryptocurrency markets show intricate and non-linear
interactions between several price-influencing elements [22]. Deep learning models can identify complicated
patterns and correlations in the data because they can represent complex functions. Human feature engineering
no longer be necessary as deep learning algorithms automatically extract relevant features from raw data.
In the Bitcoin market, where pertinent aspects could be dynamic and complicated, this is useful [23]‒[25].

Figure 2. Proposed workflow diagram

Int J Artif Intell, Vol. 14, No. 2, April 2025: 1401-1409

Int J Artif Intell ISSN: 2252-8938  1405

Table 1. Fallouts of DFT

Parameters Speed (rpm)
#Lags used 28.000000
Frequency of observations 3151.000000
Critical value (1%) -3.531219
Critical value (5%) -2.751343
Critical value (10%) -2.657257
Data type float64
Test statistic -1.342317
p-value 0.584528

4. RESULTS ANALYSIS
Cryptocurrency forecasting can assist people and organizations in evaluating and controlling the
risks brought on by price swings. Developing methods to mitigate risks is made possible by understanding
possible price fluctuations. Predictions are a common tool used by investors to maximize their bitcoin
holdings. They can modify their asset portfolios to enhance profits or reduce losses by forecasting market
fluctuations. In this work we are taking four different models from two separate fields and making the
comparison. When compared to other models, the Bi-LSTM model performs well. Figure 3 displays the
LSTM anticipated price of bitcoin, whereas Figure 4 displays the FB-Prophet expected price of bitcoin.
In a similar vein, Figures 5 and 6 illustrate the predicted prices for bitcoin by Silverkite and Bi-LSTM,
respectively.

Figure 3. LSTM predicted BTC price

Figure 4. FB-Prophet predicted BTC price

A comparative analysis of exponential smoothing method and deep learning … (Nrusingha Tripathy)
1406  ISSN: 2252-8938

Figure 5. Silverkite predicted BTC price

Figure 6. Bi-LSTM predicted BTC price

Table 2 displays the mean absolute error (MAE), mean squared error (MSE), and root mean squared
error (RMSE) scores of each model. The Bi-LSTM model has an RMSE score of 3.639 and a minimal MAE
score of 7.073 when compared to the LSTM, FB-Prophet, and Silverkite models. By utilizing the capacity to
recognize intricate trends in sequential data and to record long-term dependencies, the Bi-LSTM model
provides various potential advantages for bitcoin price prediction. Prices of cryptocurrencies frequently show
long-term trends and interdependence. As a variant of LSTM with bidirectional processing, Bi-LSTM models
are able to more accurately represent long-term dependencies in the data by capturing data from both past
and future time steps. Models' MAE, MSE, and RMSE scores are plotted in a bar diagram conspiracy in
Figure 7. Prognoses for the future can be used by cryptocurrency-related businesses to put hedging plans into
place. In order to provide some amount of risk management, this entails establishing positions to counter
possible losses from unfavorable market fluctuations. Future projections can help businesses that adopt or use
cryptocurrencies make wise decisions about financial planning, budgeting, and general company strategies.
Forecasts for the future help raise awareness and educate people about the bitcoin industry. Entrepreneurs
keep up with the dynamics of the ever-changing cryptocurrency industry by being aware of possible future
developments.

Table 2. Error score of each model

Model name MAE MSE RMSE
LSTM 14.728 26.525 5.471
FB-Prophet 8.357 15.354 3.917
Silverkite 7.662 13.948 3.834
Bi-LSTM 7.073 11.929 3.639

Int J Artif Intell, Vol. 14, No. 2, April 2025: 1401-1409

Int J Artif Intell ISSN: 2252-8938  1407

Figure 7. MAE, MSE, and RMSE result of models

5. CONCLUSION
Research and analysis of the market benefit from predictions. They offer perceptions of market
mood, possible trends, and variables affecting bitcoin pricing. Using this knowledge to inform strategic
decision-making is beneficial. Predictions can help organizations that deal with cryptocurrencies, including
exchanges and payment processors, make choices. In this work against the LSTM, FB-Prophet, and
Silverkite models, the Bi-LSTM model has a minimal MAE score of 7.073 and an RMSE score of 3.639.
This might involve risk management, price tactics, and market fluctuation preparation. Predictions can be
used by organizations that deal with cryptocurrencies, such as corporations and institutional investors,
to create hedging strategies that would lessen the risk of losses brought on by market volatility. Financial
markets are so dynamic and complicated that it is always difficult to forecast their movements, especially
those of bitcoin values. Future developments in data science and a thorough comprehension of the dynamic
cryptocurrency environment will probably be used to anticipate bitcoin prices. Scholars and experts in this
domain will persist in investigating inventive methods to augment the precision and dependability of
predictions for the bitcoin.

REFERENCES
[1] O. Kraaijeveld and J. D. Smedt, “The predictive power of public Twitter sentiment for forecasting cryptocurrency prices,” Journal
of International Financial Markets, Institutions and Money, vol. 65, 2020, doi: 10.1016/j.intfin.2020.101188.
[2] B. Chen, F. Wei, and C. Gu, “Bitcoin theft detection based on supervised machine learning algorithms,” Security and
Communication Networks, vol. 2021, 2021, doi: 10.1155/2021/6643763.
[3] S. Lahmiri and S. Bekiros, “Cryptocurrency forecasting with deep learning chaotic neural networks,” Chaos, Solitons and
Fractals, vol. 118, pp. 35–40, 2019, doi: 10.1016/j.chaos.2018.11.014.
[4] X. Sun, M. Liu, and Z. Sima, “A novel cryptocurrency price trend forecasting model based on LightGBM,” Finance Research
Letters, vol. 32, 2020, doi: 10.1016/j.frl.2018.12.032.
[5] S. Borodo, S. M. Shamsuddin, S. Hasan, and S. M. Borodo, “Big data platforms and techniques,” Indonesian Journal of Electrical
Engineering and Computer Science, vol. 1, no. 1, pp. 191–200, 2017, doi: 10.11591/ijeecs.v1.i1.pp191-200.
[6] H. I. Fawaz, G. Forestier, J. Weber, L. Idoumghar, and P. A. Muller, “Deep learning for time series classification: a review,” Data
Mining and Knowledge Discovery, vol. 33, no. 4, pp. 917–963, 2019, doi: 10.1007/s10618-019-00619-1.
[7] M. R. Cherati, A. Haeri, and S. F. Ghannadpour, “Cryptocurrency direction forecasting using deep learning algorithms,” Journal
of Statistical Computation and Simulation, vol. 91, no. 12, pp. 2475–2489, 2021, doi: 10.1080/00949655.2021.1899179.
[8] S. K. Nayak, A. K. Nayak, S. Mishra, P. Mohanty, N. Tripathy, and S. Prusty, “Improving Kui digit recognition through machine
learning and data augmentation techniques,” Indonesian Journal of Electrical Engineering and Computer Science, vol. 35, no. 2,
pp. 867–877, 2024, doi: 10.11591/ijeecs.v35.i2.pp867-877.
[9] H. Sebastião and P. Godinho, “Forecasting and trading cryptocurrencies with machine learning under changing market
conditions,” Financial Innovation, vol. 7, no. 1, 2021, doi: 10.1186/s40854-020-00217-x.
[10] M. Iqbal, M. S. Iqbal, F. H. Jaskani, K. Iqbal, and A. Hassan, “Time-series prediction of cryptocurrency market using machine
learning techniques,” EAI Endorsed Transactions on Creative Technologies, vol. 8, no. 28, 2021, doi: 10.4108/eai.7-7-2021.170286.
[11] N. Tripathy, S. Hota, and D. Mishra, “Performance analysis of bitcoin forecasting using deep learning techniques,” Indonesian Journal
of Electrical Engineering and Computer Science, vol. 31, no. 3, pp. 1515–1522, 2023, doi: 10.11591/ijeecs.v31.i3.pp1515-1522.
[12] T. Ashfaq et al., “A machine learning and blockchain based efficient fraud detection mechanism,” Sensors, vol. 22, no. 19, 2022,
doi: 10.3390/s22197162.
[13] H. Kanezashi, T. Suzumura, X. Liu, and T. Hirofuchi, “Ethereum fraud detection with heterogeneous graph neural networks,” in
International Workshop on Mining and Learning with Graphs, 2018, pp. 1–8.
[14] A. A. Amponsah, A. F. Adekoya, and B. A. Weyori, “A novel fraud detection and prevention method for healthcare claim processing
using machine learning and blockchain technology,” Decision Analytics Journal, vol. 4, 2022, doi: 10.1016/j.dajour.2022.100122.

A comparative analysis of exponential smoothing method and deep learning … (Nrusingha Tripathy)
1408  ISSN: 2252-8938

[15] N. Tripathy, S. K. Balabantaray, S. Parida, and S. K. Nayak, “Cryptocurrency fraud detection through classification techniques,”
International Journal of Electrical and Computer Engineering, vol. 14, no. 3, pp. 2918–2926, 2024, doi:
10.11591/ijece.v14i3.pp2918-2926.
[16] Y. Kumar, “AI techniques in blockchain technology for fraud detection and prevention,” in Security Engineering for Embedded
and Cyber-Physical Systems, CRC Press, 2023.
[17] V. Patel, L. Pan, and S. Rajasegarar, “Graph deep learning based anomaly detection in ethereum blockchain network,” in Network
and System Security, Cham, Switzerland: Springer, 2020, pp. 132–148, doi: 10.1007/978-3-030-65745-1_8.
[18] K. Ariya, S. Chanaim, and A. Y. Dawod, “Correlation between capital markets and cryptocurrency: impact of the coronavirus,”
International Journal of Electrical and Computer Engineering, vol. 13, no. 6, pp. 6637–6645, 2023, doi:
10.11591/ijece.v13i6.pp6637-6645.
[19] I. Nasirtafreshi, “Forecasting cryptocurrency prices using recurrent neural network and long short-term memory,” Data and
Knowledge Engineering, vol. 139, 2022, doi: 10.1016/j.datak.2022.102009.
[20] N. Tripathy, S. Hota, S. Prusty, and S. K. Nayak, “Performance analysis of deep learning techniques for time series forecasting,”
in 2023 International Conference in Advances in Power, Signal, and Information Technology (APSIT), IEEE, 2023, pp. 639–644,
doi: 10.1109/APSIT58554.2023.10201734.
[21] R. F. Ibrahim, A. M. Elian, and M. Ababneh, “Illicit account detection in the ethereum blockchain using machine learning,” in
2021 International Conference on Information Technology (ICIT), IEEE, 2021, pp. 488–493, doi:
10.1109/ICIT52682.2021.9491653.
[22] O. I. Jacinta, A. E. Omolara, M. Alawida, O. I. Abiodun, and A. Alabdultif, “Detection of Ponzi scheme on Ethereum using
machine learning algorithms,” Scientific Reports, vol. 13, no. 1, 2023, doi: 10.1038/s41598-023-45275-0.
[23] T. Hu et al., “Transaction-based classification and detection approach for Ethereum smart contract,” Information Processing and
Management, vol. 58, no. 2, 2021, doi: 10.1016/j.ipm.2020.102462.
[24] N. Tripathy, P. Satapathy, S. Hota, S. K. Nayak, and D. Mishra, “Empirical forecasting analysis of bitcoin prices: A comparison
of machine learning, deep learning, and ensemble learning models,” International Journal of Electrical and Computer
Engineering Systems, vol. 15, no. 1, pp. 21–29, 2024, doi: 10.32985/ijeces.15.1.3.
[25] R. Tan, Q. Tan, P. Zhang, and Z. Li, “Graph neural network for ethereum fraud detection,” in 2021 IEEE International
Conference on Big Knowledge (ICBK), IEEE, 2021, pp. 78–85, doi: 10.1109/ICKG52313.2021.00020.

BIOGRAPHIES OF AUTHORS

Nrusingha Tripathy received an MCA degree in computer science from

Ravenshaw University, Cuttack, Odisha, India, in 2018, and an M.Tech. degree in computer
science from Utkal University, Bhubaneswar, Odisha, India, in 2020. Currently, he is pursuing
a Ph.D. in computer science and engineering at the Institute of Technical Education and
Research (ITER) at Siksha ‘O’ Anusandhan (Deemed to be) University, Bhubaneswar, India.
His research interests include machine learning, deep learning and financial market prediction.
He has published twelve conference papers and ten journal papers. With over five years of
teaching experience. He can be contacted at email: [email protected].

Dr. Debahuti Mishra received an M.Tech. in computer science and engineering

from KIIT Deemed to be University, Bhubaneswar, India, in 2006, and a Ph.D. from Siksha
‘O’ Anusandhan (Deemed to be) University, Bhubaneswar, India, in 2011. She is currently
Professor and head of the Department of Computer Science and Engineering at ITER, Siksha
‘O’ Anusandhan (Deemed to be) University. Her research interests include data mining,
financial market prediction, and image processing. She has authored eight books and over 260
research papers. She can be contacted at email: [email protected].

Dr. Sarbeswara Hota received his MCA from the National Institute of
Technology (NIT), Rourkela, India, in 2002, and his Ph.D. in computer science and
engineering from Siksha ‘O’ Anusandhan (Deemed to be) University in 2019. He is currently
an Associate Professor in the Department of Computer Application at the same University.
His research interests include data mining, machine learning, and deep learning. He has
published 30 papers in various international journals and conferences. He can be contacted at
[email protected].

Int J Artif Intell, Vol. 14, No. 2, April 2025: 1401-1409

Int J Artif Intell ISSN: 2252-8938  1409

Dr. Mandakini Priyadarshani Behera received an M.Tech. in computer science

and engineering from Gandhi Institute of Technological Advancement, Bhubaneswar, India, in
2018, and a Ph.D. from Siksha ‘O’ Anusandhan (Deemed to be) University, Bhubaneswar,
India, in 2024. She is currently a Faculty Associate of School of Computer Engineering at
KIIT, (Deemed to be) University, Bhubaneswar, Odisha. Her research interests include
machine learning and deep learning. She has authored over 10 research papers. She can be
contacted at email: [email protected].

Gobinda Chandra Das received his MCA degree from IPSAR College Cuttack
in 2016, Odisha and an M.Tech. (computer science and engineering) form CIME College in
2019 Bhubaneswar, Odiaha. Currently working as Assistant Professor in Department of
Computer Science and Engineering KL (Deemed to be) University Vijaywada and pursuing
(Ph.D.) in KL (Deemed to) University Vijayawada (Andhrapradesh) India. He has published
2 patents, 2 journal publications, and also have 6yrs+ teaching experiences. He can be
contacted at email: [email protected].

Sasanka Sekhar Dalai received the M.Tech. degree in computer science from the
College of Engineering & Technology, BPUT Bhubaneswar, Odisha, India. He is currently
pursuing his Ph.D. in computer science and engineering at the Institute of Technical Education
and Research (I.T.E.R.) in Siksha O Anusandhan Deemed to be University, Bhubaneswar,
India, and has published one conference and two journal papers. Although, he has 9+ years of
teaching experience. His research interests include machine learning, deep learning, and image
processing. He can be contacted at email: [email protected].

Subrat Kumar Nayak received an MCA from Biju Patnaik University of

Technology, Odisha, in 2010 and an M.Tech. in computer science from Utkal University,
Bhubaneswar, in 2012. He is currently pursuing a Ph.D. in computer science and engineering
at Siksha ‘O’ Anusandhan (Deemed to be) University, Bhubaneswar, India. He has published
10 papers in international journals and conferences and qualified UGC NET in 2012. With
over six years of academic experience and three years of government experience. He can be
contacted at email: [email protected].

A comparative analysis of exponential smoothing method and deep learning … (Nrusingha Tripathy)