Cryptocurrency Price Prediction with Neural Networks of LSTM and Bayesian Optimization

Page created by Rhonda Santos
 
CONTINUE READING
RESEARCH ARTICLE
 European Journal of Business and Management Research
 www.ejbmr.org

  Cryptocurrency Price Prediction with Neural Networks of
             LSTM and Bayesian Optimization
           Ehsan Sadeghi Pour, Hossein Jafari, Ali Lashgari, Elaheh Rabiee, and Amin Ahmadisharaf

    ABSTRACT
        In this paper we present a price prediction for Bitcoin prices. The
                                                                                        Submitted : February 07, 2022
        methodology used is a hybrid artificial neural network model of Long Short-
        Term Memory and Bayesian Optimization. This is a complex model with a           Published : March 07, 2022
        high prediction power, which to our knowledge has not been applied to           ISSN: 2507-1076
        prediction of cryptocurrency prices to date. Following Charandabi and
        Kamyar (2021), we elaborate on previous methods used for prediction of          DOI: 10.24018/ejbmr.2022.7.2.1307
        cryptocurrency prices and build on their methodology. We conclude with
        detailed graphs and tables of optimization results.
                                                                                        E. S. Pour
                                                                                        Department of Electrical and Computer
        Keywords: Bayesian optimization, artificial neural networks, cryptocurrency     Engineering, Islamic Azad University,
        price prediction, long short-term memory.                                       Kashan, Iran
                                                                                        (e-mail: Ehsan61.ai@gmail.com)
                                                                                        H. Jafari
                                                                                        Master of Accounting, Allameh Amini
                                                                                        Institute     of     Higher    Education,
                                                                                        Mazandaran, Iran.
                                                                                        (email: hossein.jafari2008@gmail.com)
                                                                                        A. Lashgari*
                                                                                        Department of Economics, Kansas State
                                                                                        University, Manhattan, KS, USA.
                                                                                        (e-mail: alilashgari@ksu.edu)
                                                                                        E. Rabiee
                                                                                        Department of Economics, Kansas State
                                                                                        University, Manhattan, KS, USA.
                                                                                        (e-mail: elahehrabiee@ksu.edu)
                                                                                        A. Ahmadisharaf
                                                                                        Kansas State University, Manhattan, KS,
                                                                                        USA.
                                                                                        (e-mail: ahmadish@ksu.edu)

                                                                                       *Corresponding Author

                                                                   combination with one another to improve accuracy and
                       I. INTRODUCTION                             predictive power. Charandabi and Kamyar (2021B) present a
                                                                   survey on the literature on applications of artificial neural
   The world has gone through many changes since 2009. Just
                                                                   networks to cryptocurrency price prediction. Based on their
over the last 13 years Instagram was founded, Steve Jobs
                                                                   reports, there have been a number of hybrid and singular
passed away, a whole pandemic started, and so on. But an
                                                                   artificial neural network models implemented so far;
extremely important event that happened in 2009 hand has
                                                                   however, there are still numerous gaps to be filled in the
been changing the world since is the launch of Bitcoin. While
                                                                   literature.
the cryptocurrency was conceptualized back in 1990’s, it was
                                                                      This paper serves to fill the literature by providing a model
not relevant to the public until years after Bitcoin was first
                                                                   of Neural Networks of LSTM and Bayesian Optimization. In
implemented. The rate of increase in popularity is so high that
                                                                   Section II, we review related literature; in Section III, we
Bitcoins which barely sufficed to by a pizza in 2010, costed
                                                                   discuss the data process, in Section IV we go through
as high as $68000 in 2021.
                                                                   optimization details, and in section V we conclude.
   Financial analysts attribute the popularity of Bitcoin to its
decentralization, secure blockchain system, and control over
value. Charandabi and Kamyar (2021A) presents a thorough
                                                                                      II. LITERATURE REVIEW
survey of the history, foundation, and relevance of Bitcoin. In
the same paper, they argue on the importance and relevance            “Artificial neural networks (ANNs) are biologically
of artificial neural networks to predict prices of                 inspired computational networks. Multilayer perceptrons
cryptocurrency. Artificial neural networks may have a high         (MLPs), the ANNs most commonly used for a wide variety
predictive power and can be pertained to cryptocurrency price      of problems, are based on a supervised procedure and
data in any time horizon.                                          comprise three layers: input, hidden, and output.” (Park &
   Artificial neural network models can be used in                 Lek, 2016) In this context, artificial neural networks may be

  DOI: http://dx.doi.org/10.24018/ejbmr.2022.7.2.1307                                                  Vol 7 | Issue 2 | March 2022   20
RESEARCH ARTICLE
 European Journal of Business and Management Research
 www.ejbmr.org

powerful tools to predict growing or volatile numerical           view the data aspect of this research paper as a novel work in
values.                                                           the literature. All exchange rates are with respect to United
   Within the financial literature, artificial neural networks    States Dollars.
have been applied to prediction of stocks market indices since
decades ago. Singular networks often have a lower predictive
power than hybrid models, though they take shorter to                                   IV. OPTIMIZATION
compute. (Ghashami et al., 2021) Here we present a series of         The baseline treatment is prediction of Bitcoin prices with
price prediction for Bitcoin prices. The methodology used is      Neural Networks of LSTM and Bayesian Optimization. The
a hybrid artificial neural network model of Long Short-Term       value of optimal delays is set to 1.30, and the value of optimal
Memory and Bayesian Optimization. This complex model              training percentage (division of data to training and testing)
has a high predictive power over prediction of numerical          is set to 0.8 (Nejatian, 2022). Furthermore, to optimize speed,
values. To our knowledge, it has not been applied to the          the optimal code execution environment is the CPU. The
context of prediction of cryptocurrency prices to date            optimal drop out value is 0.5. These values are endogenous to
   “Long Short-Term Memory (LSTM) is a specific recurrent         the model and remain constant across the treatments.
neural network (RNN) architecture that was designed to               Some general deep learning parameters are exogenously
model temporal sequences and their long-range dependencies        varied across treatments. In the baseline, the maximum
more accurately than conventional RNNs.” (Sak et al., 2014)       number of training Epochs in deep learning algorithms
As described in the survey of Charandabi & Kamyar (2021B),        (maxEpoch) is set equal to 400. Following standard
LSTM is frequently a candidate for hybrid artificial neural       cryptocurrency optimization literature (Charandabi &
network models. Another optimization method, Bayesian             Kamyar, 2021C) we exogenously set it to 200 in the treatment
Optimization, is a sequential design strategy for global          run. Additionally, in the baseline, the minimum batch size of
optimization of black-box (unknown internal structure)            training Epochs in deep learning algorithms is set equal to 32.
functions (Mockus 2012). The novel hybrid structure of            Following standard cryptocurrency optimization literature
LSTM and Bayesian Optimization has been tested on stock           (Charandabi & Kamyar, 2021C) we exogenously set it to 16
market indices (Huang et al., 2018) and showed a high             in the treatment run.
predictive power, therefore, we view it as suitable for              We start reporting results by the baseline of Bitcoin prices
prediction of cryptocurrency indicators.                          (weekly from January 2020 through January 2022),
   As elaborated on in Charandabi & Kamyar (2021C), the           maximum training Epoch number of 400, and batch size of
high volatility of cryptocurrency prices roots from maximal       training Epochs in deep learning algorithms of 32. Fig. 1
trading, that stems from uncertainty. Decision-making under       depicts the input data, reporting the values for mean and
under unforeseen events has been the focus of research in the     standard deviation as well. The algorithm normalizes input
last few decades (Asadi et al., 2022). More recently,             data prior to running, in order for the base numbers to be
(Dehghani Filabadi 2019), Filabadi (2022) proposed general        feasible for computational concerns. Fig. 2 depicts the
mathematical models for addressing uncertainties in               normalized input data, reporting the values for mean and
environments where data is sensitive to prediction errors.        standard deviation as well. In the green and blue lines, Fig. 3
This approach was further extended for other applications         depicts the minimum observed objective and estimated
such as energy management (Filabadi & Azad, 2020),                minimum objective, respectively.
network management (Filabadi & Bagheri, 2021), and
inventory and healthcare (Filabadi & Mahmoudzadeh, 2022).

                            III. DATA
   We extracted data from Yahoo! Finance, a major resource
for daily data on financial indices and cryptocurrency prices.
Ghashami et al. (2021) elaborates on data gathering for
financial indicators for the purposes of time-series prediction
through artificial neural network methodology. We follow the
same system for optimization of data points among the daily
values for Open, High, Low, Close, and Adj. Close.
   Cryptocurrency data is extremely volatile by nature.
There’s a body of literature dedicated to prediction of
volatility of cryptocurrency price data, as explained
thoroughly in the survey paper Charandabi & Kamyar
(2021C). In order to avoid running into problems and
                                                                                   Fig. 1. Input data of first treatment.
retrieving weak predictions, we use weekly data, and expand
the time horizon instead. The employed data runs from
January 10, 2020, to January 10, 2022, on a weekly basis.
Most of the literature on prediction of cryptocurrency prices
employ data from a short time horizon, as explained in the
survey paper Charandabi & Kamyar (2021B). Therefore, we

  DOI: http://dx.doi.org/10.24018/ejbmr.2022.7.2.1307                                                    Vol 7 | Issue 2 | March 2022   21
RESEARCH ARTICLE
 European Journal of Business and Management Research
 www.ejbmr.org

            Fig. 2. Normalized input data of first treatment.                             Fig. 5. Error mean evaluation.

  Fig. 3. Minimum observed objective and estimated minimum objective
                               values.
                                                                                             Fig. 6. Error histogram.

   Fig. 4-7 depict technical aspects of training data. Fig. 4
shows the output evaluation and reports the rank correlation
number, which is significantly high. Fig. 5 shows the error
evaluation and reports the MSE (Mean Squared Error),
RMSE (Rooted Mean Square Error), and NRMSE
(Normalized Rooted Mean Square Error). Figure 6 shows the
error histogram, reporting Error Mean and Error Standard
Deviation numbers. Fig. 7 shows the Regression Graph
Evaluation of train data, reporting the R-squared value, that
is highly significant.

                                                                                       Fig. 7. Regression graph evaluation.

                                                                          The output results had Maximum Objective Evaluations of
                                                                       60 reached. Total time elapsed was 849.8 seconds, with a total
                                                                       function evaluation count of 60 and a total objective function
                                                                       evaluation time of 766.6 seconds. Observed objective
                                                                       function value was 0.07662, with an estimated objective
                                                                       function value of 0.15971 and a function evaluation time of
                                                                       7.8227. Also, estimated objective function value was 0.14654
                                                                       and estimated function evaluation time was 6.7355. Fig. 8- 10
                                                                       represent technical aspects of the test data. Fig. 8 depicts
                       Fig. 4. Output evaluation.
                                                                       output evaluation, and reports rank correlation value. Fig. 9
                                                                       shows the error evaluation and reports the MSE (Mean

  DOI: http://dx.doi.org/10.24018/ejbmr.2022.7.2.1307                                                       Vol 7 | Issue 2 | March 2022   22
RESEARCH ARTICLE
 European Journal of Business and Management Research
 www.ejbmr.org

Squared Error), RMSE (Rooted Mean Square Error), and               MSE (Mean Squared Error), RMSE (Rooted Mean Square
NRMSE (Normalized Rooted Mean Square Error). Fig. 10               Error), and NRMSE (Normalized Rooted Mean Square Error)
shows the error histogram, reporting Error Mean and Error          of all data. Fig. 12 depicts regression evaluation of all data,
Standard Deviation numbers.                                        reporting the R-squared value.

                 Fig. 8. Output evaluation of test data.
                                                                                      Fig. 11. Error mean evaluation.

               Fig. 9. Error mean evaluation of test data.

                                                                                   Fig. 12. Regression graph evaluation.

                                                                      Table I shows technical details of the best observed
                                                                   feasible point (above) and the best estimated feasible point
                                                                   (below), according to the models. All points are on the first
                                                                   layer.

                                                                                    TABLE I: BEST FEASIBLE POINTS
                                                                                           LSTM          Initial           Layer 2
                                                                        Number of Units
                                                                                           Layer       Learn Rate            Reg.
                                                                             68              2           0.016             0.00475
                                                                             71              2           0.021             0.00798

                                                                      The treatment run employed the same methodology and
                                                                   data, with the exception of the maximum number of training
                                                                   Epochs in deep learning algorithms (maxEpoch) set equal to
                                                                   200, and the minimum batch size of training Epochs in deep
                 Fig. 10. Error histogram of test data.            learning algorithms set equal to 16. In the green and blue
                                                                   lines, Fig. 13 depicts the minimum observed objective and
   Fig. 11 and 12 show technical details of all (i.e., train and   estimated minimum objective, respectively.
test) data. Fig. 11 depicts the error evaluation and reports the

  DOI: http://dx.doi.org/10.24018/ejbmr.2022.7.2.1307                                                   Vol 7 | Issue 2 | March 2022   23
RESEARCH ARTICLE
 European Journal of Business and Management Research
 www.ejbmr.org

 Fig. 13. Minimum observed objective and estimated minimum objective
                               values.
                                                                                             Fig. 16. Error histogram.
   Fig. 14-17 depict technical aspects of training data. Fig. 14
shows the output evaluation and reports the rank correlation
number, which is significantly high. Fig. 15 shows the error
evaluation and reports the MSE (Mean Squared Error),
RMSE (Rooted Mean Square Error), and NRMSE
(Normalized Rooted Mean Square Error). Fig. 16 shows the
error histogram, reporting Error Mean and Error Standard
Deviation numbers. Fig. 17 shows the Regression Graph
Evaluation of train data, reporting the R-squared value, that
is highly significant.

                                                                                       Fig. 17. Regression graph evaluation.

                                                                          The output results had Maximum Objective Evaluations of
                                                                       60 reached. Total time elapsed was 849.8 seconds, with a total
                                                                       function evaluation count of 60 and a total objective function
                                                                       evaluation time of 1271.6 seconds. Observed objective
                                                                       function value was 0.08878, with an estimated objective
                                                                       function value of 0.13354 and a function evaluation time of
                                                                       18.863. Also, estimated objective function value was
                                                                       0.071925 and estimated function evaluation time was 9.4717.
                      Fig. 14. Output evaluation.                      Fig. 18-20 represent technical aspects of the test data. Fig. 18
                                                                       depicts output evaluation, and reports rank correlation value.
                                                                       Fig. 19 shows the error evaluation and reports the MSE
                                                                       (Mean Squared Error), RMSE (Rooted Mean Square Error),
                                                                       and NRMSE (Normalized Rooted Mean Square Error). Fig.
                                                                       20 shows the error histogram, reporting Error Mean and Error
                                                                       Standard Deviation numbers.
                                                                          Fig. 21 and 22 show technical details of all (i.e., train and
                                                                       test) data. Fig. 21 depicts the error evaluation and reports the
                                                                       MSE (Mean Squared Error), RMSE (Rooted Mean Square
                                                                       Error), and NRMSE (Normalized Rooted Mean Square Error)
                                                                       of all data. Fig. 22 depicts regression evaluation of all data,
                                                                       reporting the R-squared value.
                                                                          Table II shows technical details of the best observed
                                                                       feasible point (above) and the best estimated feasible point
                                                                       (below), according to the models.
                    Fig. 15. Error mean evaluation.

  DOI: http://dx.doi.org/10.24018/ejbmr.2022.7.2.1307                                                       Vol 7 | Issue 2 | March 2022   24
RESEARCH ARTICLE
 European Journal of Business and Management Research
 www.ejbmr.org

                 TABLE II: BEST FEASIBLE POINTS
Number of      Number of     LSTM          Initial           Layer 2
  Layer          Units       Layer       Learn Rate            Reg.
    3             61            1           0.021            0.00881
    1             54            2           0.012            0.00889

   While the 200-16 model provides less noise in the resulting
data and filters volatilities, the 400-32 model yields a shorter
elapsed time and higher accuracy factors.

                                                                                          Fig. 21. Error mean evaluation.

                Fig. 18. Output evaluation of test data.

                                                                                       Fig. 22. Regression graph evaluation.

                                                                                              V. CONCLUSION
                                                                          In this paper, we applied a novel algorithm for prediction
                                                                       of time-series data (through hybrid artificial neural networks
                                                                       of Long Short-Term Memory and Bayesian Optimization) to
                                                                       prediction of Bitcoin data. We ran two treatments of data and
                                                                       training variables using Bitcoin prices (weekly from January
                                                                       2020 through January 2022): maximum training Epoch
              Fig. 19. Error mean evaluation of test data.             number of 400 and 200, and batch size of training Epochs in
                                                                       deep learning algorithms of 32 and 16. Results were reported
                                                                       graphically and in tables, and optimal solutions were
                                                                       comparatively shown.
                                                                          There exist many possible extensions to this paper. One
                                                                       may run the same algorithm with different data to compare
                                                                       consistency of external validity to other contexts. Other
                                                                       bitcoin prices (e.g., Ethereum) may be good candidates.
                                                                       Alternatively, within the financial data realm, stock market
                                                                       indices could be tested with the same algorithm. Furthermore,
                                                                       other hybrid models such as a hybrid of the current model
                                                                       with a GA-ANFIS artificial neural network algorithm a la
                                                                       Ghashami & Kamyar (2021) could be implemented to
                                                                       increase the accuracy. In light of the current impact and
                                                                       relevance of cryptocurrency prices, it’s essential that further
                                                                       models be tested.

                 Fig. 20. Error histogram of test data.

  DOI: http://dx.doi.org/10.24018/ejbmr.2022.7.2.1307                                                       Vol 7 | Issue 2 | March 2022   25
RESEARCH ARTICLE
 European Journal of Business and Management Research
 www.ejbmr.org

                             REFERENCES                                        Javadi, S., Maghami, A. & Hosseini, S. M. (2021) A deep learning approach
                                                                                    based on a data-driven tool for classification and prediction of
Amidi, Y., Nazari, B., Sadri, S., & Yousefi, A. (2021). Parameter Estimation        thermoelastic wave’s band structures for phononic crystals, Mechanics
     in Multiple Dynamic Synaptic Coupling Model Using Bayesian Point               of       Advanced          Materials      and       Structures,      DOI:
     Process State-Space Modeling Framework. Neural Computation,                    10.1080/15376494.2021.1983088.
     33(5), 1269-1299.                                                         Kamyar, K. (2019) “Effect of Unemployment Length on Employment
Andersen, T. G., Bollerslev, T., Diebold, F., Labys, P.: Modeling and               Expectations”. Undergraduate Economic Review.
     forecasting realized volatility. Econometrica 71, 579–625 (2003).         Kavousi-Fard, A., Mohammadi, M., Al-Sumaiti, A.S. (2021) “Effective
Charandabi, S. E., & Kamyar, K. (2021) “Using A Feed Forward Neural                 strategies of flexibility in modern distribution systems: reconfiguration,
     Network Algorithm to Predict Prices of Multiple Cryptocurrencies”.             renewable sources and plug-in electric vehicles. Flexibility in Electric
     European Journal of Business and Management Research, 6(5), 15-19.             Power Distribution Network”s, pp. 95–119. CRC Press, Boca Raton,
     https://doi.org/10.24018/ejbmr.2021.6.5.1056.                                  FL.
Charandabi, S. E., & Kamyar, K. (2021). Prediction of Cryptocurrency Price     Keshishian, M., Akbari, H., Khalighinejad, B., Herrero, J. L., Mehta, A. D.,
     Index Using Artificial Neural Networks: A Survey of the Literature.            & Mesgarani, N. (2020). Estimating and interpreting nonlinear
     European Journal of Business and Management Research, 6(6), 17-20.             receptive field of sensory neural responses with deep neural network
     https://doi.org/10.24018/ejbmr.2021.6.6.1138.                                  models. Elife, 9, e53445.
Charandabi, S. E., & Kamyar, K. (2021) “Survey of Cryptocurrency               Kharazmi, O., & Jahangard, S. (2020). A new family of lifetime distributions
     Volatility Prediction Literature Using Artificial Neural Networks,”            in terms of cumulative hazard rate function. Communications Faculty
                                                                                    of Sciences University of Ankara Series A1 Mathematics and Statistics,
     Business and Economic Research, Macrothink Institute. 12 (1).
                                                                                    69(1), 1-22.
     https://doi.org/10.5296/ber.v12i1.19301.
                                                                               Kharazmi, O., Saadatinik, A., & Jahangard, S. (2019). Odd hyperbolic cosine
Charandabi, S., & Ghanadiof, O. (2022). Evaluation of Online Markets
                                                                                    exponential-exponential (OHC-EE) distribution. Annals of Data
     Considering Trust and Resilience: A Framework for Predicting                   Science, 6(4), 765-785.
     Customer Behavior in E-Commerce. Journal of Business and                  Lei, M., and Mohammadi, M. (2021) Hybrid machine learning based energy
     Management                  Studies,            4(1),            23–33.        policy and management in the renewable-based microgrids considering
     https://doi.org/10.32996/jbms.2022.4.1.4.                                      hybrid electric vehicle charging demand. Int. J. Electr. Power Energy
Charandabi, S. E., Ghashami, F., & Kamyar, K. (2021) “US-China Tariff               Syst., 128, 106702.
     War: A Gravity Approach,” Business and Economic Research,                 Madanizadeh, S. A., and Setayesh, A. "Macroeconomic Uncertainty and
     Macrothink Institute. 11 (3). https://doi.org/10.5296/ber.v11i3.18757.         Economic Development." (2020). Working Paper.
Charandabi, S. E. (2020). Prediction of Customer Churn in Banking Industry.    Maghami, A. & Hosseini, S. M. (2021) Intelligent step-length adjustment for
     Age 18(92).                                                                    adaptive path-following in nonlinear structural mechanics based on
Chavoshi, S.F.; Mashayekhi, M. (2015) Estimating Demand Function of the             group method of data handling neural network, Mechanics of Advanced
     Fixed Line Telecommunication in Iran. Asian J. Res. Soc. Sci. Humanit.         Materials and Structures, DOI: 10.1080/15376494.2021.1880677.
     5, 275.                                                                   Mahmoudi, M. (2022). COVID Lessons: Was there any way to reduce the
Cheng, T.; Zhu, X.; Gu, X.; Yang, F.; Mohammadi, M. (2021) Stochastic               negative effect of COVID-19 on the United States economy? arXiv
     energy management and scheduling of microgrids in correlated                   preprint arXiv:2201.00274.
     environment: A deep learning-oriented approach. Sustain. Cities Soc.      Mahmoudi, M., & Ghaneei, H. (2022). Detection of structural regimes and
     69, 102856.                                                                    analyzing the impact of crude oil market on Canadian stock market:
Dehghani Filabadi, M. (2019). Robust optimization for SCED in AC-HVDC               Markov regime-switching approach. Studies in Economics and
     power systems (Master's thesis, University of Waterloo).                       Finance.
Dehghani Filabadi, M., & Bagheri, P. (2021). Robust-and-Cheap Framework        Mehrara, M., & Behradmehr, N., & Ahrari, M., & Mohaghegh, M. (2010).
     for Network Resilience: A Novel Mixed-Integer Formulation and                  Forecasting volatility of crude oil price using the GMDH neural
     Solution Method. arXiv e-prints, arXiv-2110.                                   network. Energy Economics Review, 7(25), 89-112.
Filabadi, M.D, Asadi, A.., Giahi, R., Ardakani, A.T., Azadeh, A. (2022). A     Miremadi, A., Kenar Roudi, J., & Ghanadiof, O. (2021). Evaluation on Role
     New Stochastic Model for Bus Rapid Transit Scheduling with                     of Electronic Word of Mouth (EWOM) Ads in Customers’ Emotions
     Uncertainty, Future Transportation, 1(2).                                      and Choices in E-Shops. International Journal of Industrial Marketing,
Filabadi, M. D., & Mahmoudzadeh, H. (2022). Effective Budget of                     6(1). https://doi.org/10.5296/ijim.v6i1.
     Uncertainty for Classes of Robust Optimization. INFORMS Journal on        Miremadi, A., & Ghanadiof, O. (2021). CRM Competitive Strategy in
     Optimization (in press).                                                       Financial Institutions. European Journal of Business and Management
Filabadi, M.D., (2022). A New Paradigm in Addressing Data Uncertainty:              Research, 6(3), 111-117. https://doi.org/10.24018/ejbmr.2021.6.3.867.
     Discussion and Future Research, Academia Letters, Article 4775.           Miremadi, A., Golchobian, M. M. A., & Ghanadiof, O. (2021). Requirement
     https://doi.org/10.20935/AL4775.                                               and Architecture of Organization Development. European Journal of
Filabadi, M. D., & Azad, S. P. (2020). Robust optimisation framework for            Business        and        Management         Research, 6(4),      55-64.
     SCED problem in mixed AC-HVDC power systems with wind                          https://doi.org/10.24018/ejbmr.2021.6.4.932.
     uncertainty. IET Renewable Power Generation, 14(14), 2563-2572.           Mobtahej, M., Esapour, K., Tajalli, S.Z., Mohammadi, M. (2021) Effective
Fardhosseini, M. S., Soltaninejad, M., Karji, A., Ghorbani, Z., & Ghanadiof,        demand response and GANs for optimal constraint unit commitment in
     O. (2021). Qualitative Evaluation of 5S Application Considering the            solar-tidal based microgrids. IET Renew. Power Gener. 1–11
     Experience of Electrical Construction Experts.                                 https://doi.org/10.1049/rpg2.12331.
Ghanadiof, O. (2021). Customer Loyalty and Powerful Brand in Heavy             Mockus, J. (2012). Bayesian approach to global optimization: theory and
     Machinery Industry. European Journal of Business and Management                applications. Kluwer Academic.
     Research, 6(3), 195-199. https://doi.org/10.24018/ejbmr.2021.6.3.903.     Mohaghegh, M., & A. S. Valipour, (2020). Income-dependent impacts of
Ghashami, F., & Kamyar, K. (2021) “Performance Evaluation of ANFIS and              financial development and human capital on economic growth. A non-
     GA-ANFIS for Predicting Stock Market Indices”. International                   stationary panel analysis. Theoretical and Applied Economics.
     Journal           of           Economics          and          Finance.   Mohaghegh, M., and A. S. Valipour. "Triggering Economic Growth: Trade
     https://doi.org/10.5539/ijef.v13n7p1.                                          Liberalization as the Prominent Factor in Less-developed Countries."
Ghashami, F., Kamyar, K., & Riazi, S. A. (2021) “Prediction of Stock Market         Business and Economic Research 11.2 (2021): 252-265.
     Index Using a Hybrid Technique of Artificial Neural Networks and          Mohammadi, M., Kavousi-Fard, A., Dabbaghjamanesh, M., Farughian, A.,
     Particle Swarm Optimization”. Applied Economics and Finance, 8, 1.             & Khosravi, A. (2021). Effective Management of Energy Internet in
     https://doi.org/10.11114/aef.v8i3.5195.                                        Renewable Hybrid Microgrids: A Secured Data Driven Resilient
Ghanadiof, O., Miremadi, A., & Mohammadian, M. (2021). Strategic                    Architecture. IEEE Transactions on Industrial Informatics.
     Planning and Strategic Analysis of Food Industry Using SWOT.              Nakamoto, S. (2008) “Bitcoin: A Peer-to-Peer Electronic Cash System.”
     Scientific Research Journals in Management and Social Studies, 2(23),          https://bitcoin.org/bitcoin.pdf.
     14–33.                                                                    Nazari, F. and Yan, W. Convolutional versus Dense Neural Networks:
Huang, B., Ding, Q., Sun, G., & Li, H. (2018). Stock prediction based on            Comparing the Two Neural Networks’ Performance in Predicting
     Bayesian-LSTM. Proceedings of the 2018 10th International                      Building Operational Energy Use Based on the Building Shape arXiv
     Conference       on      Machine       Learning     and     Computing.         preprint arXiv:2108.12929.
     https://doi.org/10.1145/3195106.3195170.                                  Nazemi, A.; Shaghaghi, V.; Mashayekhi, M. (2014) Market Power

   DOI: http://dx.doi.org/10.24018/ejbmr.2022.7.2.1307                                                                   Vol 7 | Issue 2 | March 2022     26
RESEARCH ARTICLE
 European Journal of Business and Management Research
 www.ejbmr.org

     Assessment of Iran’s Wholesale Electricity Market. Asian J. Res.
     Mark., 3, 172–177.
Nazemi, A.; Mashayekhi, M. Modelling Welfare Loss in Iranian Electricity
     Market. (2014) In Proceedings of the 37th IAEE International
     Conference, Energy and the Economy, New York, USA, June 2014
Nejatian, A. (2022). Time Series Prediction with Bayesian optimization,
     MATLAB Central File Exchange.
Nisar, M. U., Voghoei, S., & Ramaswamy, L. (2017, June). Caching for
     pattern matching queries in time evolving graphs: challenges and
     approaches. In 2017 IEEE 37th International Conference on
     Distributed Computing Systems (ICDCS) (pp. 2352-2357). IEEE.
Park, Y.S., & Lek, S. (2016). Artificial Neural Networks. Developments in
     Environmental           Modelling,          123–140.        Springer.
     https://doi.org/10.1016/b978-0-444-63623-2.00007-4.
Pourbemany, J., Zhu, Y., & Bettati, R. (2021). A Survey of Wearable
     Devices Pairing Based on Biometric Signals. arXivpreprint
     arXiv:2107.11685.
Pourbemany, J., Mirjalily, G., Abouei, J., & Raouf, A. H. F. (2018, May).
     Load Balanced Ad-Hoc On-Demand Routing Basedon Weighted Mean
     Queue Length Metric. In Electrical Engineering (ICEE), Iranian
     Conference on (pp. 470-475). IEEE.
Sak, H., Senior, A. W., and Beaufays, F. (2014). Long short-term memory
     recurrent neural network architectures for large scale acoustic
     modeling. In Interspeech, Google Research.
Ulstad, A., Mashayekhi, M. and Mechstroth, H. Work in Progress: Do
     Engineering Students Gain Financial Literacy Skills by Taking an
     Engineering Economy Course? (2018) American Society of
     Engineering Education (ASEE), Paper ID# 21471
Yahoo! (2022). Bitcoin USD (BTC-USD) price history & historical data.
     Yahoo!       Finance.   Retrieved     January    15,   2022,    from
     https://finance.yahoo.com/quote/BTC-USD/history/.
Yousefi, A., Amidi, Y., Nazari, B., & Eden, U. T. (2020). Assessing
     Goodness-of-Fit in Marked Point Process Models of Neural Population
     Coding via Time and Rate Rescaling. Neural Computation, 32(11),
     2145-2186.

   DOI: http://dx.doi.org/10.24018/ejbmr.2022.7.2.1307                       Vol 7 | Issue 2 | March 2022   27
You can also read