Evaluasi Potensi Search Behavior Dan Intervensi Pada Peramalan Penumpang Penerbangan Menggunakan SVR dan XGBoost Dengan Rolling Window Validation

Febrianti, Cika Rahmannia (2026) Evaluasi Potensi Search Behavior Dan Intervensi Pada Peramalan Penumpang Penerbangan Menggunakan SVR dan XGBoost Dengan Rolling Window Validation. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221081_Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Peramalan jumlah keberangkatan penumpang pesawat domestik berperan penting dalam mendukung perencanaan operasional transportasi udara, namun karakteristik data yang dipengaruhi oleh tren, pola musiman, perubahan perilaku masyarakat, serta kejadian eksternal menjadikan proses peramalan semakin kompleks. Penelitian ini bertujuan membandingkan kinerja Support Vector Regression (SVR) dan Extreme Gradient Boosting (XGBoost) dalam meramalkan jumlah keberangkatan penumpang domestik di Bandara Internasional Soekarno-Hatta menggunakan data historis, search behavior dari Google Trends, dan variabel intervensi pandemi COVID-19. Data yang digunakan berupa data bulanan periode Januari 2011 hingga Desember 2025. Seleksi variabel Google Trends dilakukan menggunakan uji Granger Causality, sedangkan pembentukan fitur historis menggunakan Partial Autocorrelation Function (PACF). Pemodelan dilakukan menggunakan tiga kombinasi variabel serta lima skenario rolling window validation berukuran 12, 24, 36, 48, dan 60 bulan. Hasil penelitian menunjukkan bahwa XGBoost memberikan performa yang lebih baik dibandingkan SVR. Model terbaik diperoleh pada XGBoost dengan kombinasi fitur historis, Google Trends, dan intervensi menggunakan rolling window 24 bulan yang menghasilkan Median MAPE sebesar 7,15% dan Median RMSE sebesar 114 ribu penumpang. Penambahan variabel Google Trends dan intervensi meningkatkan akurasi XGBoost secara lebih konsisten, sedangkan pada SVR penambahan kedua variabel tersebut tidak memberikan peningkatan akurasi. Hasil peramalan tahun 2026 menunjukkan jumlah keberangkatan penumpang diproyeksikan berada pada kisaran 1,33–1,46 juta penumpang per bulan dengan pola musiman yang tetap terjaga. Hasil penelitian menunjukkan bahwa pemanfaatan Google Trends sebagai informasi tambahan mampu meningkatkan performa XGBoost sehingga model tersebut berpotensi mendukung perencanaan operasional bandara dan pengambilan kebijakan transportasi udara.
==============================================================================================================================
Forecasting domestic air passenger departures plays an important role in supporting air transportation operational planning. However, the characteristics of passenger data, which are influenced by long-term trends, seasonal patterns, changes in public travel behavior, and external events, make the forecasting process increasingly challenging. This study aims to compare the performance of Support Vector Regression (SVR) and Extreme Gradient Boosting (XGBoost) in forecasting domestic departing passengers at Soekarno-Hatta International Airport using historical data, Google Trends search behavior, and a COVID-19 intervention variable. The study employed monthly data from January 2011 to December 2025. Google Trends variables were selected using the Granger Causality test, while historical features were constructed using the Partial Autocorrelation Function (PACF). The models were developed using three feature combinations and five rolling window validation scenarios with window sizes of 12, 24, 36, 48, and 60 months. The results indicate that XGBoost outperformed SVR. The best-performing model was XGBoost with historical features, Google Trends, and the intervention variable using a 24-month rolling window, achieving a Median MAPE of 7.15% and a Median RMSE of 114 thousand passengers. The inclusion of Google Trends and the intervention variable consistently improved the predictive performance of XGBoost, whereas these additional variables did not significantly improve the performance of SVR. Forecasting results for 2026 project domestic departing passengers to range from approximately 1,33 to 1,46 million passengers per month while maintaining recurring seasonal patterns. These findings demonstrate that incorporating Google Trends as an additional source of information enhances the performance of XGBoost, making it a promising approach for supporting airport operational planning and air transportation policy-making.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Peramalan Penumpang, Rolling Window Validation, Search Behavior, Support Vector Regression, Extreme Gradient Boosting, Passenger Forecasting
Subjects:	H Social Sciences > HA Statistics > HA30.3 Time-series analysis Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	Faculty of Mathematics, Computation, and Data Science > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Cika Rahmannia Febrianti
Date Deposited:	24 Jul 2026 21:53
Last Modified:	24 Jul 2026 21:53
URI:	http://repository.its.ac.id/id/eprint/137326

Actions (login required)

View Item