Stock Trend Prediction Using a Multivariate Hybrid Attention-Based BiLSTM with 1D Convolution Model

The accurate prediction of stock trends in Bangladesh presents unique challenges due to the lack of tailored advanced models capable of capturing the country’s economic intricacies. This study proposes a novel hybrid deep learning model, the Multivariate Hybrid Attention-based BiLSTM with 1D Convolution (MHA-BiLSTM-CNN), specifically designed to address these challenges. The proposed model integrates Bidirectional LSTM layers to capture long-term and short term dependencies, an attention mechanism to focus on critical temporal features, and a 1D convolutional layer to efficiently extract local patterns within the data. The MHA-BiLSTM-CNN model is evaluated against four other advanced models: Attention-based BiLSTM, Attention-based GRU, Attention-based BiGRU, and Attention-based CNN. The proposed MHA-BiLSTM-CNN model achieves superior predictive performance, with 94.32% accuracy, 1.49% MAPE, 0.92 R™, and 0.96 Pearson’s Correlation, outperforming Attention-based BiLSTM, GRU, BiGRU, and CNN in comparative analysis. This hybrid approach significantly advances stock market forecasting, offering valuable insights for financial analysis and facilitating data driven decision making in Bangladesh’s financial markets.