Stock Price Prediction Using LSTMSeq2Seq: A Deep Learning Approach with Advanced Feature Engineering

This paper explores stock price prediction in the context of the Bangladesh stock market using advanced deep learning methodologies, specifically the LSTMSeq2Seq model. Traditional analytical techniques often fail to capture the complex, non-linear patterns inherent in financial data. To address this limitation, our research leverages the strengths of the LSTM-Seq2Seq architecture, which is tailored to the unique dynamics of the Bangladesh market. A robust feature engineering process is employed, utilizing 67 diverse features, including market trends, stock-specific indicators, fundamental and technical variables, and key macroeconomic factors such as gold prices and foreign exchange reserves. The dataset encompasses historical stock market data from 2012 to 2022, and extensive data preprocessing is performed to handle inconsistencies. Hyperparameter optimization is conducted using grid search to ensure model performance. The LSTMSeq2Seq model is evaluated based on key metrics such as MAE, MAPE, MSE, and the R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> score, demonstrating superior predictive capabilities compared to traditional methods. This research not only advances the application of deep learning in financial forecasting but also provides valuable insights into stock price prediction in emerging markets, addressing ethical concerns to mitigate risks like market manipulation.