Application of SARIMA Model with Additive Outliers for Predicting Train Passenger at Kediri Station

Abstract

Forecasting the number of train passengers is an important component in transportation system management to support operational planning and decision-making. Kediri Station, as a regional transit hub, exhibits highly dynamic daily passenger patterns with frequent fluctuations. The passenger time series data from this station shows occasional extreme spikes, particularly during holidays, long weekends, and promotional periods. These spikes can be characterized as Additive Outliers (AO), which may distort the underlying data structure and reduce the reliability of forecasting models if not properly addressed. This study aims to improve short-term forecasting accuracy of daily train passengers at Kediri Station by applying the Seasonal Autoregressive Integrated Moving Average (SARIMA) model with Additive Outlier (AO) handling. The analysis uses secondary daily passenger data from January 2024 to September 2025. AO effects are incorporated into the SARIMA framework to reduce bias caused by extreme observations. The results show that the SARIMA-AO model yields improved forecasting performance compared to the standard SARIMA model. The SARIMA-AO model achieves a Mean Squared Error (MSE) of 30,913 and a Mean Absolute Percentage Error (MAPE) of 13%, while the standard SARIMA model produces an MSE of 33,965 and a MAPE of 14%. Although the improvement is relatively modest, the results indicate that explicit handling of additive outliers can enhance forecast accuracy for daily passenger demand at Kediri Station. This approach provides preliminary evidence that AO-aware SARIMA modeling may be useful for supporting short-term operational planning in railway services.