Bitcoin Price Trend Forecasting in a Dynamic Market: A Superior CNN-LSTM Hybrid Approach

Abstract

Predicting Bitcoin’s price movements is a difficult task for researchers, investors, and traders. While many studies focus on daily price predictions, short-term trends within a day are not well studied. In this research, we use advanced deep learning models—CNN (model 1), LSTM (model 2), and a hybrid CNN model (model 3)—to improve the accuracy of Bitcoin trend predictions. We compare these models to find the best approach. This study explores the effectiveness of different deep learning architectures in predicting Bitcoin price trends. The results highlight significant improvements in model accuracy and reduction in loss across three models: CNN, LSTM, and a Hybrid approach. Model 1 (CNN) achieved an accuracy of 86.00% with a loss of 0.3646, demonstrating strong predictive capabilities but slightly higher error. Model 2 (LSTM) performed similarly, also achieving 86.00% accuracy but with a lower loss of 0.3443, indicating a more stable learning process. However, Model 3 (Hybrid) outperformed both, achieving the highest accuracy of 87.50% with the lowest loss of 0.3129, suggesting improved pattern recognition and better generalization. The reduced loss in Model 3 signifies fewer errors, making it more reliable for financial market forecasting. These findings emphasize the impact of architectural choices and optimization techniques on predictive performance. Overall, the Hybrid model proves to be the most effective, providing a promising approach for enhancing Bitcoin market trend predictions and financial decision-making.