PREDICTING CRYPTOCURRENCY PRICE MOVEMENTS USING AI: INTEGRATING TECHNICAL ANALYSIS AND CYCLICAL THEORIES WITH MACHINE LEARNING MODELS

The increasing complexity of forecasting cryptocurrency price movements, especially Bitcoin, necessitates the development of innovative approaches that address high market volatility, dependency on investor sentiment, limited historical data, and regulatory uncertainty. This article aims to provide a robust forecasting framework that integrates advanced AI models and synthetic data generation techniques. The relevance of this research lies in the growing influence of cryptocurrencies in the global financial market, where accurate price prediction remains a key challenge. The proposed methodology combines the generative capabilities of a WGAN with the predictive strength of a BiLSTM network. The WGAN model was specifically designed to generate synthetic time series data that replicate real BTC price patterns. To enhance data quality and realism, the model incorporates additional market indicators, including trading volume, RSI, and volatility metrics, as well as time-conditioned parameters that reflect market cycles. The synthetic dataset generated by WGAN closely mirrors real market dynamics, providing a solid foundation for predictive modeling. The BiLSTM model was trained using both real and synthetic data, enabling a comparative analysis of predictive performance. Evaluation metrics such as MAE, RMSE, MAPE, and the R² score were employed to assess model accuracy. The results demonstrated a significant improvement in predictive performance when the BiLSTM model was trained on synthetic data, with the R² score reaching 0.97 and error rates considerably lower than those observed in models trained solely on real data. This outcome highlights the potential of synthetic data to overcome the limitations imposed by scarce historical data and improve the generalization capability of forecasting models. Furthermore, the study analyzed the impact of incorporating technical indicators into the prediction process. The inclusion of RSI led to enhanced predictive accuracy, while other indicators, such as MACD, support and resistance levels, and Fibonacci retracement, showed varying degrees of influence. The practical value of this study lies in its demonstration that the hybrid integration of WGAN and BiLSTM provides a scalable and adaptable framework for forecasting cryptocurrency prices. By effectively compensating for limited historical data and adapting to the unique characteristics of the crypto market, this approach offers practical benefits for institutional investors, financial analysts, and trading platforms seeking to optimize investment strategies and manage risks in highly volatile environments.