Pemodelan Klaim Agregat PT Asuransi Umum Bumida dengan Metode XGBoost Berbasis Hasil Dimensionality Reduction Deep Autoencoder dan Deep Variational Autoencoder

Fatoni, Ivan (2025) Pemodelan Klaim Agregat PT Asuransi Umum Bumida dengan Metode XGBoost Berbasis Hasil Dimensionality Reduction Deep Autoencoder dan Deep Variational Autoencoder. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5006211062-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Klaim agregat merupakan komponen penting dalam analisis risiko dan penentuan kebijakan pada industri asuransi. Namun, kompleksitas struktur data klaim dengan berbagai variabel dan hubungan nonlinier menimbulkan tantangan dalam proses pemodelan prediktif. Penelitian ini bertujuan untuk mengevaluasi dan membandingkan efektivitas dua metode reduksi dimensi, yaitu Deep Autoencoder (DAE) dan Deep Variational Autoencoder (DVAE), dalam meningkatkan akurasi model klaim agregat menggunakan algoritma utama Extreme Gradient Boosting (XGBoost). Data yang digunakan berasal dari PT. Asuransi Umum Bumiputera Muda 1967 dan telah melalui tahap pra-pemrosesan mencakup cleaning data, feature engineering, normalisasi, serta tambahan cross validation menggunakan K-Fold. Tiga pendekatan pemodelan diuji, yaitu XGBoost tanpa reduksi dimensi, XGBoost dengan DAE, dan XGBoost dengan DVAE. Evaluasi kinerja dilakukan berdasarkan tiga metrik utama, yaitu Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), dan Root Mean Squared Error (RMSE). Hasil pengujian menunjukkan bahwa model XGBoost dengan reduksi dimensi menggunakan Deep Autoencoder memberikan performa terbaik dengan nilai MAPE sebesar 0.1202, MAE sebesar 0.00059, dan RMSE sebesar 0.00159. Sebaliknya, penggunaan DVAE menunjukkan performa terburuk, diduga karena kompleksitas arsitektur dan distribusi latent space yang tidak sesuai dengan karakteristik data klaim penelitian ini. Kesimpulannya, penerapan Deep Autoencoder sebagai metode pra-pemrosesan terbukti efektif dalam meningkatkan akurasi model prediktif klaim agregat berbasis XGBoost. Penelitian ini diharapkan dapat memberikan kontribusi bagi pengembangan model berbasis deep learning dalam manajemen risiko dan estimasi cadangan klaim di industri asuransi.
===============================================================================================================================
Aggregate claims are an important component in risk analysis and policy determination in the insurance industry. However, the complexity of the claims data structure with multiple variables and nonlinear relationships poses challenges in the predictive modeling process. This study aims to evaluate and compare the effectiveness of two dimension reduction methods, namely Deep Autoencoder (DAE) and Deep Variational Autoencoder (DVAE), in improving the accuracy of aggregate claims prediction using Extreme Gradient Boosting (XGBoost) algorithm. The data used comes from PT Asuransi Umum Bumiputera Muda 1967 and has gone through a pre-processing stage including data cleaning, feature engineering, normalization, and additional cross validation using K-Fold. Three modeling approaches were tested, namely XGBoost without dimensionality reduction, XGBoost with DAE, and XGBoost with DVAE. Performance evaluation is performed based on three main metrics, namely Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE). The test results show that the XGBoost model with dimension reduction using Deep Autoencoder provides the best performance with a MAPE value of 0.1202, MAE of 0.00059, and RMSE of 0.00159. In contrast, the use of DVAE showed the worst performance, presumably due to the complexity of the architecture and the latent space distribution that did not match the characteristics of this study's claim data. In conclusion, the application of Deep Autoencoder as a pre-processing method proved effective in improving the accuracy of the XGBoost-based aggregate claims predictive model. This research is expected to contribute to the development of deep learning-based models in risk management and claims reserve estimation in the insurance industry.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Aggregate Claims, Autoencoder, Dimensionality Reduction, Variational Autoencoder, XGBoost, Autoencoder, Klaim Agregat, Reduksi Dimensi, Variational Autoencoder, XGBoost
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA278.5 Principal components analysis. Factor analysis. Correspondence analysis (Statistics) Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) Q Science > QA Mathematics > QA76.9.D33 Data compression (Computer science)
Divisions:	Faculty of Mathematics, Computation, and Data Science > Actuaria > 94203-(S1) Undergraduate Thesis
Depositing User:	Ivan Fatoni
Date Deposited:	30 Jul 2025 07:27
Last Modified:	30 Jul 2025 07:27
URI:	http://repository.its.ac.id/id/eprint/123462

Actions (login required)

View Item