Pemodelan Frekuensi Klaim Kendaraan Bermotor Perusahaan Z Menggunakan Zero-Inflated Poisson Berbasis Categorical Boosting

Fadhel, Muhammad Nabel (2026) Pemodelan Frekuensi Klaim Kendaraan Bermotor Perusahaan Z Menggunakan Zero-Inflated Poisson Berbasis Categorical Boosting. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5006221002-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Asuransi kendaraan bermotor merupakan salah satu lini bisnis utama pada industri asuransi umum di Indonesia. Data frekuensi klaim asuransi kendaraan bermotor umumnya memiliki karakteristik excess zero dan overdispersi sehingga regresi Poisson standar menjadi tidak tepat. Penelitian ini bertujuan memodelkan frekuensi klaim kendaraan bermotor Perusahaan Z menggunakan pendekatan Zero-Inflated Poisson berbasis Generalized Linear Model atau ZIP GLM sebagai baseline dan Zero-Inflated Poisson berbasis Categorical Boosting atau ZIP Catboost sebagai model machine learning. Data yang digunakan berjumlah 179.714 observasi periode 2019 sampai 2023 dengan 13 variabel prediktor. Tahapan analisis meliputi pembersihan data, statistika deskriptif, uji multikolinieritas menggunakan Cramer's V, serta pembentukan model dengan skenario variabel penuh dan variabel terseleksi. Performa keempat model dievaluasi menggunakan Log-Likelihood, Deviance, Pseudo R-squared, dan Uji Vuong. Hasil penelitian menunjukkan model ZIP Catboost Full menjadi model terbaik dengan nilai Pseudo R-squared sebesar 0,8359, mengungguli ZIP GLM pada seluruh metrik dan terbukti signifikan melalui Uji Vuong. Interpretasi model terbaik menggunakan Feature Importance dan SHAP menunjukkan bahwa riwayat klaim atau NOC dan penyebab kerugian atau Cause of Loss merupakan variabel paling berpengaruh terhadap frekuensi klaim.
=======================================================================================================================================
Motor vehicle insurance is one of the main business lines in the general insurance industry in Indonesia. Claim frequency data in motor vehicle insurance generally exhibit excess zero and overdispersion, making the standard Poisson regression inappropriate. This study aims to model the motor vehicle claim frequency of Company Z using the Zero-Inflated Poisson based on Generalized Linear Model or ZIP GLM as a baseline and the Zero-Inflated Poisson based on Categorical Boosting or ZIP Catboost as a machine learning model. The data consist of 179.714 observations from 2019 to 2023 with 13 predictor variables. The analysis includes data cleaning, descriptive statistics, multicollinearity testing using Cramer'sV, and model development under full and selected variable scenarios. The performance of the four models is evaluated using Log-Likelihood, Deviance, Pseudo R-squared, and the Vuong test. The results show that the ZIP Catboost Full model is the best model with a Pseudo R-squared of 0,8359, outperforming ZIP GLM across all metrics and proven significant through the Vuong test. Interpretation of the best model using Feature Importance and SHAP reveals that claim history or NOC and cause of loss are the most influential variables on claim frequency.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Asuransi Kendaraan Bermotor, Catboost, Excess Zero, Frekuensi klaim, Zero-Inflated Poisson.
Subjects:	Q Science Q Science > QA Mathematics Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry) Q Science > QA Mathematics > QA278.2 Regression Analysis. Logistic regression
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Actuaria > 94203-(S1) Undergraduate Thesis
Depositing User:	Muhammad Nabel Fadhel
Date Deposited:	16 Jul 2026 02:41
Last Modified:	16 Jul 2026 02:41
URI:	http://repository.its.ac.id/id/eprint/135105

Actions (login required)

View Item