TLACP: A Hybrid Deep Learning Model for Stock Market Prediction

Complexity, non-linearity, and extreme volatility make stock market prediction difficult. Using traditional methods like SVR, MLP, and LSTM for time series analysis and machine learning can be challenging for handling high-dimensional, non-linear data. The Transformer-LSTM-Attention-CNN with Learnable Positional Embedding (TLACP) hybrid deep learning model was proposed to address these challenges and improve stock market predictions. The TLACP model enhances performance in handling complex stock market data by combining Transformers' global attention mechanism, LSTM's temporal data processing, CNN's feature extraction, and learnable positional embedding techniques. Daily Chinese A-share trade data was used to test and compare the model to standard models. Statistical metrics like RMSE, MAE, MAPE, and R2 showed superior performance in all experiments with the TLACP model. Ablation experiments verified each TLACP model component's efficacy and requirement. This study shows that the TLACP hybrid deep learning model can forecast financial data, providing stock finance with fresh insights and solutions.