پیش‌بینی قیمت رمز ارز بیت کوین با استفاده از شبکه عصبی مصنوعی بهینه‌سازی شده توسط الگوریتم‌های بهینه‌سازی فراابتکاری
محورهای موضوعی : فصلنامه اقتصاد محاسباتیآیدین ابوطالبی 1 , کامبیز پیکارجو 2 , ابراهیم رضایی 3 , رحیم خانیزاد 4
1 - علوم تحقیقات، تهران، ایران
2 - علوم تحقیقات، تهران، ایران
3 - علوم تحقیقات، تهران، ایران
4 - علم و صنعت، تهران، ایران
کلید واژه: پیش‌بینی قیمت رمز ارز, شبکه عصبی مصنوعی, بیت کوین, الگوریتم‌های بهینه‌سازی فراابتکاری,
چکیده مقاله :
رمز ارزها، ارزهای دیجیتالی مبتنی بر فناوری بلاک‌چین هستند که با استفاده از تکنولوژی رمزنگاری، امنیت و کنترل ایجاد واحدهای جدید را تأمین می‌کنند و به صورت مستقل از بانک‌ها و دولت‌ها عمل می‌کنند. بیت‌کوین، اتریوم و لایت‌کوین از معروف‌ترین رمز ارزها هستند که برای تبادل ارزش و سرمایه‌گذاری استفاده می‌شوند. پیش‌بینی قیمت این ارزها با تحلیل داده‌های تاریخی و عوامل مؤثر بر بازار انجام می‌شود تا به سرمایه‌گذاران در تصمیم‌گیری بهتر کمک کند. برای موفقیت در بازارهای مالی، تخصص و آگاهی ضروری است؛ لذا سرمایه‌گذاران همواره به دنبال روش‌هایی برای پیش‌بینی دقیق‌تر قیمت رمزارزها، به‌ویژه بیت کوین، هستند. با پیشرفت کامپیوترها و افزایش قدرت پردازشی آن‌ها، روشهای هوش مصنوعی مانند شبکه‌های عصبی مصنوعی به ابزارهایی برای پیش‌بینی دقیق‌تر قیمت رمز ارزها تبدیل شده‌اند. در این پژوهش، به پیش‌بینی قیمت رمز ارز بیت کوین در بازه‌های زمانی کوتاه‌مدت (10روزه) و بلندمدت (30 روزه) با استفاده از شبکه عصبی مصنوعی بهینه سازی شده توسط شش الگوریتم بهینه‌سازی فراابتکاری به نام‌های الگوریتم بهینه ساز سیاسی (PO)، الگوریتم سلسله مراتب رتبه در شرکت (HBO)، الگوریتم بهینه ساز رنگ‌های تصادفی (SPO)، الگوریتم ساخت اهرام جیزه (GPC)، الگوریتم بهینه ساز شاهین آتشین (FHO) و الگوریتم بهینه ساز روباه (FOX) پرداخته شده و نتایج این شش حالت با یکدیگر مقایسه شده است. نتایج نشان می‌دهد که در بازه زمانی کوتاه مدت 10 روزه و بلند مدت 30 روزه، شبکه عصبی بهینه‌سازی شده توسط الگوریتم GPC به ترتیب با ضریب تعیین 88/0 و 72/0 بهتر از سایر الگوریتم‌ها بوده است.
Extended Abstract
Purpose
Digital currencies, or cryptocurrencies, are electronic assets that utilize blockchain technology to ensure the security and transparency of transactions. These currencies operate in a decentralized manner and are not controlled by any central authority. Bitcoin, as the first and most well-known cryptocurrency, was created in 2009 by an anonymous individual or group under the name Satoshi Nakamoto. Subsequently, other cryptocurrencies such as Ethereum, Ripple, and Litecoin were developed.
Transactions using cryptocurrencies are often faster and less expensive than traditional banking methods. These digital currencies have become a popular investment option, but their extreme volatility can increase investment risks. In many countries, clear regulations regarding the use of cryptocurrencies have not yet been established, which may pose legal challenges for users. The application of artificial intelligence and machine learning models, particularly artificial neural networks, is crucial for predicting cryptocurrency prices. These digital currencies are accompanied by vast amounts of data, and analyzing them manually can be challenging. AI can process and analyze this extensive data. Artificial neural networks can identify complex, nonlinear patterns in data that may not be detectable by traditional methods. Machine learning models, using advanced algorithms, can enhance the accuracy of cryptocurrency price predictions and assist investors in making better decisions. These models can perform analyses faster than traditional methods, which is highly significant in the volatile cryptocurrency markets.
Methodology
Artificial neural networks consist of numerous processing units called neurons that operate in coordination, similar to human brain neurons, to solve problems. The connections between these neurons determine the network's performance. In these networks, a data structure is designed through programming to act like neurons, and then connections between the neurons are established, followed by the application of a training algorithm to train the network. Artificial neural networks learn similarly to humans through examples.
Artificial neural networks have the ability to adapt to rapid market changes and can identify new patterns. Utilizing AI can help reduce errors caused by emotional decision-making or human biases. These models can simultaneously analyze the impact of various factors, such as economic news, regulatory changes, and social trends. Machine learning models, with access to more data and continuous feedback, can continuously improve their performance. The more precise information and deeper analyses provided by these technologies can serve as powerful tools for investors, minimizing risks. The real-time implementation of these models is possible, which is a significant advantage for active traders.
Based on the aforementioned considerations in this research, the focus has been on predicting Bitcoin prices over short-term (10-day) and long-term (30-day) periods. For this purpose, an artificial neural network optimized by six metaheuristic optimization algorithms has been employed. These algorithms include the Political Optimizer (PO), Hierarchical Rank-based Optimization (HBO), Stochastic Colors Optimizer (SPO), Giza Pyramid Construction (GPC), Firefly Optimizer (FHO), and Fox Optimizer (FOX). Finally, the results obtained from these algorithms have been compared.
In this study, Bitcoin price data were collected and examined over a 13-year period, from July 13, 2010, to December 6, 2023. This dataset includes 4,895 samples and six variables. The variables include the opening price of Bitcoin, the highest price, the lowest price, trading volume, and market value as inputs, with the final price of Bitcoin considered as the output.
Finding
This research is designed as an applied study, aiming to create a framework for predicting Bitcoin prices over short-term (10-day) and long-term (30-day) periods using the machine learning method known as artificial neural networks, optimized with the six metaheuristic algorithms: PO, HBO, SPO, GPC, FHO, and FOX.
The steps of the work are as follows: Initially, in MATLAB software, the necessary data for modeling is preprocessed, and missing or unavailable values are removed. Then, to reduce the impact of data scaling on the results, data normalization is performed within the range of 0 to 1, as entering raw data can decrease the model's speed and accuracy. Subsequently, 80 percent of the initial Bitcoin price data is divided into training data, while the remaining 20 percent is allocated as test data.
Afterward, using the training dataset, the artificial neural network (ANN) is trained with the mentioned optimization algorithms. In the next stage, the optimized neural network models are tested using the test dataset. Finally, the results and accuracy of various models in predicting Bitcoin prices in both training and testing scenarios are compared. Evaluation criteria include the coefficient of determination (R²), root mean square error (RMSE), mean square error (MSE), mean absolute error (MAE), relative square error (RSE), and explained variance (EVS). All modeling steps are conducted in MATLAB R2023b.
Conclusion
The results of this research indicate that in the short-term (10-day) period, the neural network optimized with the GPC algorithm achieved a coefficient of determination of 0.88, and in the long-term (30-day) period, this network again performed best compared to other algorithms with a coefficient of determination of 0.7.
Aggarwal, A., Gupta, I., Garg, N., & Goel, A. (2019). Deep learning approach to determine the impact of socio economic factors on bitcoin price prediction. Paper presented at the 2019 twelfth international conference on contemporary computing (IC3).
Akba, F., Medeni, I. T., Guzel, M. S., & Askerzade, I. J. I. A. (2021). Manipulator detection in cryptocurrency markets based on forecasting anomalies. 9, 108819-108831.
Askari, Q., Saeed, M., & Younas, I. J. E. S. w. A. (2020). Heap-based optimizer inspired by corporate rank hierarchy for global optimization. 161, 113702.
Askari, Q., Younas, I., & Saeed, M. J. K.-b. s. (2020). Political Optimizer: A novel socio-inspired meta-heuristic for global optimization. 195, 105709.
Azizi, M., Talatahari, S., & Gandomi, A. H. J. A. I. R. (2023). Fire Hawk Optimizer: A novel metaheuristic algorithm. 56(1), 287-363.
Chhajer, P., Shah, M., & Kshirsagar, A. J. D. A. J. (2022). The applications of artificial neural networks, support vector machines, and long–short term memory for stock market prediction. 2, 100015.
Choithani, T., Chowdhury, A., Patel, S., Patel, P., Patel, D., & Shah, M. J. A. o. D. S. (2024). A comprehensive study of artificial intelligence and cybersecurity on bitcoin, crypto currency and banking system. 11(1), 103-135.
Dutta, A., Kumar, S., Basu, M. J. J. o. r., & management, f. (2020). A gated recurrent unit approach to bitcoin price prediction. 13(2), 23.
Engelbrecht, A. P. (2007). Computational intelligence: an introduction: John Wiley & Sons.
Harifi, S., Mohammadzadeh, J., Khalilian, M., & Ebrahimnejad, S. J. E. I. (2021). Giza Pyramids Construction: an ancient-inspired metaheuristic algorithm for optimization. 14(4), 1743-1761.
Hassani, H., Huang, X., & Silva, E. (2018). Big-Crypto: Big Data, Blockchain and Cryptocurrency. 2(4), 34.
Hornik, K., Stinchcombe, M., & White, H. J. N. n. (1989). Multilayer feedforward networks are universal approximators. 2(5), 359-366.
Hussein, N., & Abdulazeez, A. M. J. I. J. o. C. S. (2024). Bitcoin Price Prediction Using Hybrid LSTM-GRU Models. 13(1).
Katsiampa, P. J. E. l. (2017). Volatility estimation for Bitcoin: A comparison of GARCH models. 158, 3-6.
Kaveh, A., Talatahari, S., & Khodadadi, N. J. E. w. C. (2022). Stochastic paint optimizer: theory and application in civil engineering. 1-32.
Koker, T. E., Koutmos, D. J. J. o. R., & Management, F. (2020). Cryptocurrency trading using machine learning. 13(8), 178.
McNally, S., Roche, J., & Caton, S. (2018). Predicting the price of bitcoin using machine learning. Paper presented at the 2018 26th euromicro international conference on parallel, distributed and network-based processing (PDP).
Mohammed, H., & Rashid, T. J. A. I. (2023). FOX: a FOX-inspired optimization algorithm. 53(1), 1030-1050.
Shahbazi, Z., & Byun, Y.-C. J. S. (2022). Knowledge discovery on cryptocurrency exchange rate prediction using machine learning pipelines. 22(5), 1740.
Shi, Y., Tian, Y., Kou, G., Peng, Y., & Li, J. (2011). Optimization based data mining: theory and applications: Springer Science & Business Media.
Sin, E., & Wang, L. (2017). Bitcoin price prediction using ensembles of neural networks. Paper presented at the 2017 13th International conference on natural computation, fuzzy systems and knowledge discovery (ICNC-FSKD).
Smith, M. (1993). Neural networks for statistical modeling: Thomson Learning.
Tayib, H., & Abdulazeez, A. M. J. I. J. o. C. S. (2024). A Review of Bitcoin Price Prediction Based on Deep Learning Algorithms. 13(2).
Umer, Q., Li, J.-W., Ashraf, M. R., Bashir, R. N., & Ghous, H. J. I. A. (2023). Ensemble deep learning based prediction of fraudulent Cryptocurrency transactions.
Xiong, L., & Lu, Y. (2017). Hybrid ARIMA-BPNN model for time series prediction of the Chinese stock market. Paper presented at the 2017 3rd International conference on information management (ICIM).
ابوصالح, م., کاوه خلیلی, د., فرشید, ع., & سهیلا, س. (1400). پیش بینی قیمت بیت کوین با استفاده از مدل ترکیبی ARIMA و یادگیری عمیق. مطالعات مدیریت صنعتی, 19(61), 125-146. الهام, ع., فاطمه, خ., & بهزاد, ه. (1401). طراحی مدلی مشارکتی با استفاده از الگوریتم های هوش مصنوعی به منظور پیش بینی قیمت بیت کوین. Paper presented at the دومین کنفرانس مکانیک،برق ،مهندسی هوافضا و علوم مهندسی. https://civilica.com/doc/1448408/
بشیری, م., & پاریاب, س. J. ا. ک. (2020). پیش بینی قیمت بیت کوین با استفاده از الگوریتم های یادگیری ماشین. 10(34 و 35 (پاییز و زمستان 99)), 1-13.
سعید, ن., & امیر, ن. (1399). استفاده از گراف معامله بیت کوین و الگوریتم های یادگیری ماشین برای پیش بینی قیمت بیت کوین. Paper presented at the چهارمین کنفرانس ملی کامپیوتر، فناوری اطلاعات و کاربردهای هوش مصنوعی. https://civilica.com/doc/1238763/
سکینه صیادی, ن., علی اسماعیل, ز., & محمدرضا, ر. (1402). ارائه مدل پیش بینی بازدهی بیت کوین با استفاده از روش هیبریدی یادگیری عمیق - الگوریتم تجزیه سیگنال (CEEMD- DL). اقتصاد مالی, 17(62), 217-238.
کیارش, آ., & عباس, ک. (1393). بررسی روشهای سنتی و مدرن پیش بینی قیمت سهام در بازار بورس و معرفی روش بهینه. Paper presented at the اولین همایش ملی پژوهش های مهندسی رایانه. https://civilica.com/doc/347317/
محمدباقر, م. (1381). مبانی شبکه‌های عصبی هوش محاسباتی (Vol. 0): دانشگاه صنعتی امیرکبیر (پلی‌تکنیک تهران)، واحدتفرش، مرکز نشر پروفسور.
