A Hybrid LSTM Model with Feature Selection and ACO-Based Hyperparameter Optimization for Stock Price Prediction in the Banking Sector

Time series analysis plays a crucial role in decision-making and policy planning, as it enables the forecasting of future values based on historical data. This study explores the integration of feature selection and the Long ShortTerm Memory (LSTM) model to enhance the accuracy of stock price predictions. The stock data of U.S. Bancorp (USB), Bank of America (BAC), and Citigroup (C) from the U.S. banking sector are preprocessed, technical indicators are computed, and key features are selected using the Random Forest (RF) method. The LSTM model is trained to learn sequential data relationships, and hyperparameter optimization is performed using Ant Colony Optimization (ACO). The model’s performance is evaluated using error metrics such as RMSE, MAE, and MAPE, revealing that proper feature selection significantly enhances prediction accuracy. The findings indicate that hyperparameter optimization with ACO, particularly when combined with effective feature selection, substantially improves stock price prediction accuracy. The results show that BAC exhibits the best predictive performance with the lowest error rates, while USB and C demonstrate relatively higher prediction errors. However, the high R2 values across all banks confirm the overall robustness of the proposed model.