Financial Market Prediction using Deep Learning Models: A Comparative Analysis of Trading Strategies

This dissertation investigates the effectiveness of three deep learning architectures- LSTM, CNN, and Transformer models —in predicting financial market movements using Bitcoin historical data. The study compares model performance using MAE, RMSE, and R² metrics, and develops a CNN-based trading strategy that achieved a 39.6% return on a backtested portfolio. Results demonstrate that the CNN model outperforms both LSTM and Transformer models, achieving the lowest error rates (MAE: 0.0156, R²: 0.9909).