Forecasting Volume Penumpang Harian KRL Yogyakarta–Solo Menggunakan SARIMA, LSTM, dan SARIMA-LSTM

Abstract

Passenger volume forecasting in public transportation is an important aspect of supporting data-driven operational decision-making. The Yogyakarta–Solo Commuter Rail is a strategic public transportation mode that supports interregional mobility. Fluctuations in passenger volume influenced by daily patterns, weekends, national holidays, and collective leave periods require forecasting models capable of capturing seasonal patterns and possible nonlinear changes in time series data. This study aims to compare the performance of Seasonal Autoregressive Integrated Moving Average (SARIMA), Long Short-Term Memory (LSTM), and hybrid SARIMA-LSTM models in forecasting the daily passenger volume of the Yogyakarta–Solo Commuter Rail. The dataset consists of daily passenger volume data from January to December 2025. The research stages include data preprocessing, calendar-based feature engineering, chronological training and testing data splitting, SARIMA modeling, LSTM modeling, residual modeling using LSTM, and performance evaluation using MAE, RMSE, and MAPE. The results show that the SARIMA model obtained an MAE of 2644.81, an RMSE of 3299.78, and a MAPE of 10.56%. The LSTM model obtained an MAE of 1977.50, an RMSE of 2528.75, and a MAPE of 7.28%. Meanwhile, the hybrid SARIMA-LSTM model achieved an MAE of 2634.78, an RMSE of 3294.24, and a MAPE of 10.52%. Based on these results, the LSTM model achieved the best forecasting performance compared to SARIMA and hybrid SARIMA-LSTM. The hybrid SARIMA-LSTM model provided only a slight improvement over SARIMA but did not outperform the LSTM model. These findings indicate that forecasting model selection should consider dataset characteristics, residual patterns, and the adequacy of historical data. Future research is recommended to use longer historical data and incorporate relevant external variables to improve forecasting accuracy.