Pemodelan Nonparametrik Untuk Intensitas Spatial Point Process Menggunakan XGBoostPP: Studi Kasus Gempa Bumi Di Sumatra

Abbad, Muhammad Akmal (2026) Pemodelan Nonparametrik Untuk Intensitas Spatial Point Process Menggunakan XGBoostPP: Studi Kasus Gempa Bumi Di Sumatra. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221087-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Gempa bumi merupakan salah satu bencana alam yang menyebabkan banyak kerugian untuk Indonesia karena berdampak pada berbagai aspek kehidupan. Pulau Sumatra termasuk wilayah paling rawan dengan aktivitas tektonik yang sangat kompleks, sehingga diperlukan suatu pemodelan bahaya seismik yang akurat dan fleksibel. Pemodelan intensitas gempa bumi umumnya dilakukan menggunakan spatial point process terhadap fungsi kovariat seperti faktor geologis. Penelitian ini bertujuan untuk mengimplementasikan model XGBoostPP, suatu metode nonparametrik berbasis gradient boosting yang diadaptasi dalam kerangka point process yang fleksibel dan mampu memodelkan banyak kovariat, untuk mengestimasi fungsi intensitas kejadian gempa bumi Sumatra. Data yang digunakan adalah kejadian gempa bumi dengan magnitudo ≥5 pada 1 Januari 2004 hingga 30 September 2025. Kovariat yang digunakan meliputi faktor geologi (jarak terdekat ke patahan, zona subduksi, dan gunung berapi) serta geometri lempeng subduksi (kedalaman, kemiringan, jurus). Hasil analisis menunjukkan bahwa distribusi gempa bumi di Sumatra bersifat tidak homogen dan cenderung clustering lemah pada jarak kecil. Selain itu, model parametrik seperti model log-linear tidak mampu menangkap hubungan kovariat terhadap intensitas secara memadai, sehingga diperlukan pendekatan nonparametrik. Evaluasi performa model menunjukkan bahwa XGBoostPP dengan Poisson loss memberikan kinerja yang sedikit lebih baik dibandingkan dynamic weighted Poisson loss, dengan nilai log-likelihood sebesar 2935,674, meskipun perbedaan kinerjanya tidak signifikan. Temuan ini mengindikasikan bahwa pola pengelompokan kejadian gempa bumi telah dijelaskan secara efektif oleh kovariat dan fungsi intensitas model, sehingga pembobotan dinamis tidak memberikan peningkatan kinerja yang signifikan. Model terbaik juga mampu memetakan risiko gempa bumi secara spasial dan mengidentifikasi kovariat dominan, terutama kedalaman dan kemiringan lempeng subduksi serta jarak ke zona subduksi, yang relevan untuk penilaian bahaya seismik di Sumatra.
=======================================================================================================================================
Earthquakes are one of the natural disasters that cause significant losses for Indonesia because they impact various aspects of life. The island of Sumatra is one of the most vulnerable regions due to its highly complex tectonic activity, requiring accurate and flexible seismic hazard modeling. Earthquake intensity modeling is generally performed using spatial point processes on covariate functions such as geological factors. This study aims to implement the XGBoostPP model, a nonparametric gradient boosting method adapted within a flexible point process framework capable of modeling many covariates, to estimate the intensity function of Sumatra earthquake events. The data used are earthquake events with a magnitude of ≥5 from January 1, 2004, to September 30, 2025. The covariates used include geological factors (closest distance to faults, subduction zones, and volcanoes) and subduction plate geometry (depth, slope, strike). The analysis results show that the distribution of earthquakes in Sumatra is non-homogeneous and tends to cluster weakly at short distances.. In addition, parametric models such as the log-linear model are unable to adequately capture the covariate relationship with intensity, thus requiring a nonparametric approach. Model performance evaluation shows that XGBoostPP with Poisson loss provides slightly better performance than dynamic weighted Poisson loss, with a log-likelihood value of 2935,674, although the difference in performance is not significant. These findings indicate that the clustering pattern of earthquake events has been effectively explained by the covariates and intensity function of the model, so that dynamic weighting does not provide a significant improvement in performance. The best model is also capable of spatially mapping earthquake risk and identifying dominant covariates, particularly the depth and dip of the subduction plate and the distance to the subduction zone, which are relevant for seismic hazard assessment in Sumatra.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Bencana Alam, Estimasi Fungsi Intensitas Gempa, Machine Learning, Spatial Point Process, XGBoostPP. ====================================================================== Natural Disasters, Earthquake Intensity Function Estimation, Machine Learning, Spatial Point Process, XGBoostPP.
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QE Geology > QE538.8 Earthquakes. Seismology
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Akmal Abbad
Date Deposited:	29 Jan 2026 08:42
Last Modified:	29 Jan 2026 08:42
URI:	http://repository.its.ac.id/id/eprint/131169

Actions (login required)

View Item