Integrating Genetic Algorithms with LSTM for Improved Public Transportation Passenger Forecasting in Thailand

Abstract This study presents a novel forecasting framework that integrates Genetic Algorithms (GA) with Long Short-Term Memory (LSTM) networks to enhance the prediction accuracy of passenger volumes in Thailand's public transportation systems. The unique contribution of this research lies in leveraging GA for automatic hyperparameter optimization of LSTM models, improving performance over conventional methods. The dataset comprises weekly aggregated passenger counts from road and rail modes between January 2020 and August 2023. The proposed GA-enhanced LSTM model (LSTM+GA) is evaluated using four metrics: Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Median Absolute Percentage Error (MdAPE). Results demonstrate that LSTM+GA outperforms baseline models including Autoregressive Integrated Moving Average (ARIMA), Facebook Prophet (FBProphet), and standard LSTM, achieving a MAPE of 4.04 for road and 5.86 for rail datasets. These findings suggest that the proposed model offers a practical and scalable tool for transport planners and public agencies seeking to optimize forecasting strategies and decision-making.