Performance Analysis of LSTM, Bidirectional LSTM (BiLSTM), Gated Recurrent Unit (GRU), Convolutional LSTM (ConvLSTM) and LSTM with Attention in Stock Market Prediction

Predicting the stock market is a complex and challenging endeavor because financial time series are inherently volatile and nonlinear. Recurrent neural networks (RNNs) and their advanced variants have shown significant promise in modeling sequential data. This study evaluates and contrasts the effectiveness of five deep learning models Classic LSTM, Bidirectional LSTM (BiLSTM), GRU, Convolutional LSTM (ConvLSTM), and LSTM with Attention Mechanism in forecasting stock market closing prices. Our evaluation employs historical stock index data, with model performance quantified using standard regression metrics—MSE, RMSE, MAE, MAPE, and R². Our results show that GRU consistently outperforms other models in capturing intricate temporal patterns, Results show that LSTM with Attention Mecha-nism achieves the highest accuracy (MSE: 0.000370 MAE: 0.014482, RMSE: 0.019242, MAPE: 6.676856 R2: 0.873357) while GRU offers the best balance of accuracy and efficiency. These findings provide actionable insights for selecting appropriate RNN models for financial forecasting applications.