Lawrence, Lawrence (2024) Perbandingan Random Survival Forests Menggunakan Splitting Berbasis Axis dan Accelerated Oblique untuk Prediksi Risiko Klaim Asuransi Penyakit Kritis. Other thesis, Institut Teknologi Sepuluh Nopember.
Text
5006201026-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 April 2026. Download (3MB) | Request a copy |
Abstract
Asuransi penyakit kritis adalah produk asuransi yang memberikan perlindungan dari risiko berbagai penyakit kritis. Sebagai pihak yang akan menanggung risiko tersebut, baik perusahaan asuransi maupun reasuransi perlu melakukan estimasi risiko klaim dan mengetahui faktorfaktor yang mempengaruhinya. Analisis terhadap risiko klaim dapat dilakukan dengan metode analisis survival. Random survival forests (RSF) adalah metode pohon gabungan yang mampu menganalisis data survival dan robust terhadap data outlier. Penelitian RSF umumnya menggunakan splitting berbasis axis yang membagi data secara rekursif menggunakan satu prediktor. Performa prediksi RSF dapat ditingkatkan ketika menggunakan splitting berbasis accelerated oblique yang membagi data menggunakan kombinasi linear dari prediktor berdasarkan koefisien regresi Cox menggunakan satu kali iterasi Newton Raphson scoring. Pada penelitian ini, RSF dengan splitting berbasis axis dan accelerated oblique akan dibandingkan untuk prediksi risiko klaim asuransi penyakit kritis di PT. Asuransi X berdasarkan C-index dan variable importance. Performa prediksi RSF juga akan dibandingkan ketika digunakan pada data pada periode COVID-19. Kemudian, dilakukan prediksi risiko klaim asuransi penyakit kritis berdasarkan nilai cumulative hazard menggunakan metode RSF terbaik. Hasil analisis menunjukkan RSF dengan splitting berbasis axis mengidentifikasi usia, premi, dan status merokok sebagai faktor yang berkontribusi terhadap prediksi. Sedangkan, RSF dengan splitting berbasis accelerated oblique mengidentifikasi usia dan premi sebagai faktor yang berkontribusi. Splitting berbasis accelerated oblique menghasilkan RSF yang lebih baik dengan peningkatan C-index sekitar 3% hingga 4% dan waktu komputasi 8 hingga 12 kali lebih cepat dibandingkan splitting berbasis axis. Splitting berbasis accelerated oblique bekerja lebih baik pada data tanpa periode COVID-19 meskipun tidak ada penurunan performa yang drastis ketika digunakan pada data dengan periode COVID-19. Splitting berbasis axis bekerja lebih baik pada data dengan periode COVID-19 dibandingkan pada data tanpa periode COVID19. Rata-rata risiko klaim berdasarkan cumulative hazard seluruh polis sangat kecil dengan nilai sekitar 0,008 pada akhir waktu prediksi.
=================================================================================================================================
Critical illness insurance is an insurance product which provides protection from several critical illness risks. As the entity that will cover these risks, either insurance or reinsurance company must estimate the claim risk and understand the factors that affect it. The analysis of claim risk can be carried out using survival analysis. Random survival forests (RSF) is an ensemble-tree method that is able to analyze survival data and is robust towards outliers.. RSF research generally uses axis-based splitting that recursively split data using a single predictor. RSF prediction performance can be improved when using accelerated oblique splitting that splits data by using a linear combination of predictors based on the Cox regression coefficients using a single Newton Raphson scoring iteration. In this study, RSF with axis-based and accelerated oblique splitting will be compared to predict critical illness insurance claim risk at PT. Asuransi X based on C-index and variable importance. RSF prediction performance will also be compared when used on data with COVID-19 period. Subsequently, critical illness insurance claim risk will be predicted based on cumulative hazard using the best RSF method. Analysis results showed that axis-based splitting RSF identified age, premium, and smoking status as contributing factors for prediction. Meanwhile, accelerated oblique splitting RSF identified age and premium as the contributing factors. Accelerated oblique splitting resulted in a better RSF with around 3% to 4% C-index improvement and took 8 to 12 times faster to compute than axis-based splitting. Accelerated oblique splitting performed better on data without COVID-19 period although there was no drastic decrease in performance when used on data with COVID-19 period. Axis-based splitting performed better on data with COVID-19 compared to data with COVID-19 period. Claim risk average based on all policy cumulative hazard is miniscule ranging around 0,008 at the end of prediction time.
Item Type: | Thesis (Other) |
---|---|
Uncontrolled Keywords: | Accelerated Oblique Splitting, Axis-Based Splitting, Claim, Survival Analysis, Analisis Survival, Klaim, Random Survival Forests, Splitting Berbasis Accelerated Oblique, Splitting Berbasis Axis |
Subjects: | Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry) |
Divisions: | Faculty of Science and Data Analytics (SCIENTICS) > Actuaria > 94203-(S1) Undergraduate Thesis |
Depositing User: | Lawrence . |
Date Deposited: | 18 Jan 2024 01:00 |
Last Modified: | 18 Jan 2024 01:00 |
URI: | http://repository.its.ac.id/id/eprint/105539 |
Actions (login required)
View Item |