PREDIKSI PENUMPANG LRT JAKARTA MENGGUNAKAN SARIMAX DAN XGBOOST DENGAN EFEK KALENDER

Muhammad Hafiz Fazli (Study Program in Statistics and Data Science, Institut Pertanian Bogor, Jl. Meranti Wing 22 Level 4 Kampus IPB Darmaga, Darmaga, Bogor, Jawa Barat, Indonesia 16680)
M. Taqy Abiyu Dzakwan (Study Program in Statistics and Data Science - School of Data Science, Mathematics, and Informatics, IPB University, Bogor, Indonesia)
Nada Ardelia (Study Program in Statistics and Data Science - School of Data Science, Mathematics, and Informatics, IPB University, Bogor, Indonesia)
Gemala Aleida Fitri (Study Program in Statistics and Data Science - School of Data Science, Mathematics, and Informatics, IPB University, Bogor, Indonesia)
Akbar Rizki (Study Program in Statistics and Data Science - School of Data Science, Mathematics, and Informatics, IPB University, Bogor, Indonesia)
Windi Pangesti (Study Program in Statistics and Data Science - School of Data Science, Mathematics, and Informatics, IPB University, Bogor, Indonesia)

Jurnal Gaussian
Vol 15, No 1 (2026): Jurnal Gaussian

The daily passenger volume of Jakarta’s LRT fluctuates significantly due to weekly seasonality and calendar variations, making accurate forecasting important for operational planning and decision-making. This study aims to determine the most effective model for forecasting daily passenger demand by comparing the SARIMAX and XGBoost methods on transportation data characterized by strong seasonal patterns and external influences. SARIMAX was selected because it models seasonal and autoregressive structures alongside exogenous variables, while XGBoost captures nonlinear relationships between temporal features and external factors. The dataset covers the period from 1 January 2024 to 31 August 2025 and includes variables such as weekends, national holidays, and special events. Model evaluation was conducted using walk-forward cross-validation and hyperparameter tuning. The results show that the SARIMAX(1,0,1)(0,1,1)7 model achieved the best performance, with a validation MAPE of 11.26% and a test MAPE of 8.64%, outperforming XGBoost. SARIMAX also reproduced weekly fluctuation patterns more consistently, indicating that it is more suitable for forecasting transportation demand with strong seasonal characteristics and relatively stable external influences.