Research Article


DOI :10.26650/ekoist.2024.40.1411482   IUP :10.26650/ekoist.2024.40.1411482    Full Text (PDF)

Performance Comparison of Genetic and Machine Learning Algorithms in Crypto Markets

Berna Yaman ŞahinSema Ulutürk Akman

The evolution of statistical methodologies for research analysis has notably contributed to the diversification of analytical and predictive techniques. Notably, machine learning, which leverages mathematical and statistical approaches to draw meaningful inferences from data, has made remarkable strides in artificial intelligence,generating predictions based on these inferences.Encompassing a spectrum of algorithms that transform datasets into models, machine learning emerges as a cornerstone in analytical and predictive processes. Herein, weproduce high-accuracy buying and selling signals in the cryptomarket—a market that continuously operates 24 h a day. This is achieved by integrating MACD (Moving Average Convergence Divergence) parameters optimized with a genetic algorithm specific tothe cryptocurrency market, machinelearning methods, and technical analysis indicators. Contextually, we compared the performances of different machine learning algorithms. Using genetic algorithm optimization, we identified the most suitable model.Results underscore the enhanced profitability of trades executed with optimized MACD parameterscompared with those executed using nonoptimizedMACD parameters. Themodel performed optimallyon the LTCUSDT pair. Notably,the deep learning algorithm exhibitedbetter profitability in the LTCUSDT pair.However, its effectiveness in generating profits in the ADAUSDT pair was somewhere limited;this can be attributed to the high volatility, instability, and rapid response of the cryptomarket to current news, whether positive or negative. Therefore,the developed model fits different cryptocurrency pairs to varying degrees. 

DOI :10.26650/ekoist.2024.40.1411482   IUP :10.26650/ekoist.2024.40.1411482    Full Text (PDF)

Kripto Piyasalarında Genetik ve Makine Öğrenmesi Algoritmaları ile Performans Karşılaştırması

Berna Yaman ŞahinSema Ulutürk Akman

Araştırma alanında kullanılan analiz yöntemlerine yönelik istatistiksel metotların gelişmesi, analiz ve öngörü tekniklerinin çeşitlenmesine önemli bir katkıda bulunmuştur. Bu kapsamda, özellikle matematiksel ve istatistiksel metodolojiler kullanarak verilerden anlamlı çıkarımlar yapabilen ve bu çıkarımları kullanarak birtakım tahminlerde bulunan makine öğrenmesi, yapay zekâ alanında önemli bir gelişme kaydetmiştir. Makine öğrenmesi, bir veri setini modele dönüştüren çeşitli algoritmaları kapsar ve bu algoritmalar, analiz ve öngörü süreçlerinde temel bir disiplin olarak öne çıkmaktadır. Bu çalışma, kripto para piyasasında genetik algoritma ile optimize edilmiş MACD parametrelerini, makine öğrenmesi yöntemleri ve teknik analiz göstergeleri ile birleştirerek, 24 saat sürekli işlem gören kripto piyasasında yüksek doğrulukta alım ve satım sinyalleri üretmeyi amaçlamaktadır. Bu bağlamda farklı makine öğrenmesi algoritmalarının performansları karşılaştırılmış ve genetik algoritma ile optimize edilerek en uygun modele ulaşılmaya çalışılmıştır. Sonuç olarak, optimize edilmiş MACD parametreleri kullanılarak yapılan işlemlerin, optimize edilmemiş MACD parametreleriyle yapılanlardan daha iyi kârlılık sağladığı gözlemlenmiştir. Modelin, LTCUSDT çiftinde daha iyi performans sergilediği sonucuna varılmıştır. Özellikle derin öğrenme algoritmasının LTCUSDT paritesinde daha iyi kâr elde edebildiği ancak, modelin ADAUSDT çiftinde kâr elde edemediği görülmüştür. Bunun sebebi de kripto piyasasının volatilitesinin yüksek, istikrarsız ve güncel haberlere olumlu/olumsuz çok hızlı tepki vermesinden kaynaklanmaktadır. Buradan yola çıkarak geliştirilen modelin farklı kripto para çiftlerine farklı derecelerde uyduğu sonucuna varılmıştır.


EXTENDED ABSTRACT


Machine learning algorithms are increasingly being embraced for their easy implementation and effective outcomes, particularly in financial predictions. In financial applications, these algorithms have emerged as potent tools forinformation processing and analysis, particularly in trading strategies,price predictions, and portfolio management. Deep learning algorithms excel in creating high-performance classification models, especially in complex datasets. Decision tree algorithms are suitable for solving classification and regression problems, whereas genetic algorithms have emerged as optimal methods for addressing complex problems, optimization, and modeling systems related to randomness.

We aim to produce high-accuracy buying and selling signals in the cryptomarket, operating continuously 24 ha day. The approach combines MACD parameters, optimized with a genetic algorithm in the cryptocurrency market, machine learning methods, and technical analysis indicators. Machine learning algorithms are essential tools that transform datasets into models. We compared the performances of different machine learning algorithms. 

Furthermore, we examine the results obtained by optimizing the performance of these algorithms using genetic algorithms. The data collection process commenced by obtaining candlestick data for ADAUSDT and LTCUSDT assets through the API of Binance Crypto Exchange.A 1-h candlestick chart was used for the analysis, spanning November 9, 2021, to November 22, 2022. Each candlestick data item includes information such as transaction time, opening, closing, high values, low values, and trading volume. The candlestickdata were separated into optimization and test data. The optimization data were created by adding CMF, RSI, Z-SCORE, ATR, BOP, CMO, and CC INDICATOR signals during the strategy generation phase. Next, thisdataset was dividedinto machine learning training and test data. Long and short trading signals were established based on MACD and Signal line crossovers, and a decision column was added.

During the optimization phase, the parameters of the MACD indicator were optimized using a genetic algorithm. The MACD indicator uses the default parameters of 12, 26, and 9. Employing the NSGA-II genetic algorithm, the optimal parameter values for the MACD indicator were calculated over 25 cycles. Subsequently, optimized MACD parameters were identified for ADAUSDT and LTCUSDT based on the test results. Backtesting was performed using the optimized MACD parameters to obtain gain and profitability ratios. Upon examining the accuracy rates of various machine learning algorithms,deep learning exhibitedthe highest accuracy rate.

The optimized MACD parameters, training data of the most suitable machine learning algorithm, and the test data were consolidated.The test data were evaluated by backtesting, employing optimized MACD parameters and trained machine learning algorithms;this process revealedthe profit rate and profitability. The results indicate a high gain ratio and profitability in the LTCUSDT asset when optimized MACD parameters and a deep learning algorithm were employed. However, for the ADAUSDT asset, the combination failed to generate a profit. This highlights that the success of certain combinations of assets and strategies may vary depending on market conditions. Therefore, for effective implementation of the proposed model, each asset’sstrategy needs careful evaluation.

In backtests focusing on ADAUSDT and LTCUSDT cryptocurrency pairs, long–short trades using optimized MACD parameters yieldedhigher profits thantrades executed with nonoptimized settings. The default MACD parameters typically involve three parameters: a fast period of 12, a slow period of 26, and a correction period of 9. However,the averages obtained from the optimization process deviatednoticeably from these standard values,suggesting that investors engaged in cryptoasset trading maybenefit from an indicator responding rapidly to market dynamics. 

Additionally, owingto the volatile nature of cryptoassets, the applicationof machine learning algorithms poses challenges in accurately determining profits and losses before ensuring a clear data separation. This underscores the risk of exposure to sudden price fluctuations in cryptoassets. In conclusion, this study found that long–short trades using optimized MACD parameters do not always yield optimal results. However, profitability can be enhancedby leveraging machine learning algorithms to filter out data leading to losses. Notably, the developed model demonstrated better compatibility with the LTCUSDT pair than the ADAUSDT, showcasing a profit of 93.55% in the LTCUSDT pair. These results underscorethe potential advantagesof integrating machine learning and technical analysis in the cryptocurrency market.


PDF View

References

  • Aguirre, A. A. A., Medina, R. A. R., & Mendez, N. D. D. (2020). Machine learning applied in the stock market through the Moving Average Convergence Divergence (MACD) indicator. Investment Management & Financial Innovations, 17(4), 44. google scholar
  • Chatterjee, S., Sarkar, S., Dey, N., Ashour, A. S., & Sen, S. (2018). Hybrid non-dominated sorting genetic algorithm: II-neural network approach. In Advancements in Applied Metaheuristic Computing (pp. 264-286). IGI Global. google scholar
  • Choudhry, R., & Garg, K. (2008). A hybrid machine learning system for stock market forecasting. International Journal of Computer and Information Engineering, 2(3), 689-692. google scholar
  • Çalış, A., Kayapınar, S., & Çetinyokuş, T. (2014). Veri Madenciliğinde Karar Ağaci Algoritmalari ile Bilgisayar ve İnternet Güvenliği Üzerine Bir Uygulama. Endüstri Mühendisliği, 25(3), 2-19. google scholar
  • Demirel, A. C., & Hazar, A. (2021). Kripto para değerlerine dayanılarak BİST 100 endeks hareketi tahmininde destek vektör makineleri uygulaması. Başkent Üniversitesi Ticari Bilimler Fakültesi Dergisi, 5(1), 27-35. google scholar
  • Gupta, B., Rawat, A., Jain, A., Arora, A. ve Dhami, N. (2017). Veri madenciliğinde sınıflandırmaya yönelik çeşitli karar ağacı algoritmalarının analizi. Uluslararası Bilgisayar Uygulamaları Dergisi, 163 (8), 15-19. google scholar
  • Holland, J. H. (1992). Adaptation in natural and artificial systems: an introductory analysis with applications to biology, control, and artificial intelligence. google scholar
  • Kara, Y., Boyacioglu, M. A., & Baykan, Ö. K. (2011). Predicting direction of stock price index movement using artificial neural networks and support vector machines: The sample of the Istanbul Stock Exchange. Expert systems with Applications, 38(5), 5311-5319. google scholar
  • Karadağ, K. (2022). Hibrit derin öğrenme modelleri ile hisse senedi fiyat tahmini (Master’s thesis, Trakya Üniversitesi Sosyal Bilimler Enstitüsü). google scholar
  • Kim, H., & Shin, K. (2007). A hybrid approach based on neural networks and genetic algorithms for detecting temporal patterns in stock markets. Applied Soft Computing, 7(2), 569-576. doi:10.1016/j.asoc.2006.03.004 google scholar
  • Kumar, M., Husain, M., Upreti, N., & Gupta, D. Genetic algorithm: review and application. Available at SSRN 3529843 (2010). google scholar
  • Latha, R. S., Sreekanth, G. R., Suganthe, R. C., Geetha, M., Selvaraj, R. E., Balaji, S., ... & Ponnusamy, P. P. (2022, January). Stock Movement Prediction using KNN Machine Learning Algorithm. In 2022 International Conference on Computer Communication and Informatics (ICCCI) (pp. 1-5). IEEE. google scholar
  • Odabaşı, M. B., & Toklu, M. C. Yapay Sinir Ağları ve Derin Öğrenme Algoritmalarının Kripto Para Fiyat Tahmininde Karşılaştırmalı Analizi. Journal of Intelligent Systems: Theory and Applications, 6(2), 96-107. google scholar
  • Olorunnimbe, K., & Viktor, H. (2023) Deep Learning in The Stock Market—A Systematic Survey of Practice, Backtesting, and Applications. Artif Intell Rev, 56, 2057-2109. https://doi.org/10.1007/s10462-022-10226-0 google scholar
  • Pabuçcu, H., Ongan, S., & Ongan, A. (2023). Forecasting the movements of Bitcoin prices: an application of machine learning algorithms. arXiv preprint arXiv:2303.04642. google scholar
  • Pai, P. F., &Wei,W. R. (2007, December). Predicting movement directions of stock index futures by support vector models with data preprocessing. In 2007 IEEE International Conference on Industrial Engineering and Engineering Management (pp. 169-173). IEEE. google scholar
  • Sel, A. (2020). Pandemi sürecinde altın fiyatları ile kripto para ilişkisinin makine öğrenme metotları ile incelenmesi. İstatistik ve Uygulamalı Bilimler Dergisi, 1(2), 85-98. google scholar
  • Singh, S., & Gupta, P. (2014). Comparative study ID3, cart and C4. 5 decision tree algorithm: a survey. International Journal of Advanced Information Science and Technology (UAIST), 27(27), 97-103. google scholar
  • Şenol, D., & Denizhan, B. (2023). Kripto Para Değerinin Yapay Sinir Ağları ile Tahmini. Endüstri Mühendisliği, 34(1), 42-69. google scholar
  • Yavuz, S., & Deveci, M. (2012). İstatiksel normalizasyon tekniklerinin yapay sinir ağin performansina etkisi. Erciyes Üniversitesi İktisadi ve İdari Bilimler Fakültesi Dergisi, (40), 167-187. google scholar
  • Yusoff, Y., Ngadiman, M. S., & Zain, A. M. (2011). Overview of NSGA-II for optimizing machining process parameters. Procedia Engineering, 15, 3978-3983. google scholar

Citations

Copy and paste a formatted citation or use one of the options to export in your chosen format


EXPORT



APA

Yaman Şahin, B., & Ulutürk Akman, S. (2024). Performance Comparison of Genetic and Machine Learning Algorithms in Crypto Markets. EKOIST Journal of Econometrics and Statistics, 0(40), 151-164. https://doi.org/10.26650/ekoist.2024.40.1411482


AMA

Yaman Şahin B, Ulutürk Akman S. Performance Comparison of Genetic and Machine Learning Algorithms in Crypto Markets. EKOIST Journal of Econometrics and Statistics. 2024;0(40):151-164. https://doi.org/10.26650/ekoist.2024.40.1411482


ABNT

Yaman Şahin, B.; Ulutürk Akman, S. Performance Comparison of Genetic and Machine Learning Algorithms in Crypto Markets. EKOIST Journal of Econometrics and Statistics, [Publisher Location], v. 0, n. 40, p. 151-164, 2024.


Chicago: Author-Date Style

Yaman Şahin, Berna, and Sema Ulutürk Akman. 2024. “Performance Comparison of Genetic and Machine Learning Algorithms in Crypto Markets.” EKOIST Journal of Econometrics and Statistics 0, no. 40: 151-164. https://doi.org/10.26650/ekoist.2024.40.1411482


Chicago: Humanities Style

Yaman Şahin, Berna, and Sema Ulutürk Akman. Performance Comparison of Genetic and Machine Learning Algorithms in Crypto Markets.” EKOIST Journal of Econometrics and Statistics 0, no. 40 (Dec. 2024): 151-164. https://doi.org/10.26650/ekoist.2024.40.1411482


Harvard: Australian Style

Yaman Şahin, B & Ulutürk Akman, S 2024, 'Performance Comparison of Genetic and Machine Learning Algorithms in Crypto Markets', EKOIST Journal of Econometrics and Statistics, vol. 0, no. 40, pp. 151-164, viewed 23 Dec. 2024, https://doi.org/10.26650/ekoist.2024.40.1411482


Harvard: Author-Date Style

Yaman Şahin, B. and Ulutürk Akman, S. (2024) ‘Performance Comparison of Genetic and Machine Learning Algorithms in Crypto Markets’, EKOIST Journal of Econometrics and Statistics, 0(40), pp. 151-164. https://doi.org/10.26650/ekoist.2024.40.1411482 (23 Dec. 2024).


MLA

Yaman Şahin, Berna, and Sema Ulutürk Akman. Performance Comparison of Genetic and Machine Learning Algorithms in Crypto Markets.” EKOIST Journal of Econometrics and Statistics, vol. 0, no. 40, 2024, pp. 151-164. [Database Container], https://doi.org/10.26650/ekoist.2024.40.1411482


Vancouver

Yaman Şahin B, Ulutürk Akman S. Performance Comparison of Genetic and Machine Learning Algorithms in Crypto Markets. EKOIST Journal of Econometrics and Statistics [Internet]. 23 Dec. 2024 [cited 23 Dec. 2024];0(40):151-164. Available from: https://doi.org/10.26650/ekoist.2024.40.1411482 doi: 10.26650/ekoist.2024.40.1411482


ISNAD

Yaman Şahin, Berna - Ulutürk Akman, Sema. Performance Comparison of Genetic and Machine Learning Algorithms in Crypto Markets”. EKOIST Journal of Econometrics and Statistics 0/40 (Dec. 2024): 151-164. https://doi.org/10.26650/ekoist.2024.40.1411482



TIMELINE


Submitted28.12.2023
Accepted29.02.2024
Published Online15.05.2024

LICENCE


Attribution-NonCommercial (CC BY-NC)

This license lets others remix, tweak, and build upon your work non-commercially, and although their new works must also acknowledge you and be non-commercial, they don’t have to license their derivative works on the same terms.


SHARE




Istanbul University Press aims to contribute to the dissemination of ever growing scientific knowledge through publication of high quality scientific journals and books in accordance with the international publishing standards and ethics. Istanbul University Press follows an open access, non-commercial, scholarly publishing.