Perbandingan Penggunaan Least Absolute Shrinkage And Selection Operator (LASSO) pada Random Forest dan Neural Network dalam Pemodelan Lapse Polis Asuransi

Siregar, Agnes Putri Victoria (2026) Perbandingan Penggunaan Least Absolute Shrinkage And Selection Operator (LASSO) pada Random Forest dan Neural Network dalam Pemodelan Lapse Polis Asuransi. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5006221041_Undergraduate_Thesis.pdf Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Lapse merupakan pembatalan polis asuransi jiwa sebelum tanggal jatuh tempo yang dapat berdampak signifikan terhadap profitabilitas perusahaan asuransi. Perilaku lapse oleh pemegang polis disebabkan oleh faktor yang beragam. Penelitian ini bertujuan untuk membandingkan penggunaan Least Absolute Shrinkage and Selection Operator (LASSO) pada algoritma Random Forest dan Neural Network dalam memprediksi lapse polis asuransi jiwa. Data yang digunakan merupakan data karakteristik polis dan pemegang polis dari perusahaan asuransi jiwa di Amerika Serikat periode Januari 2012 hingga November 2017 dengan total 16.140 data. Pada Neural Network, LASSO berfungsi sebagai regularisasi untuk mengurangi bobot, melakukan neuron pruning, dan feature selection secara serentak, sedangkan pada Random Forest, LASSO digunakan untuk mengganti mekanisme voting dengan memberikan bobot pada setiap pohon keputusan. Melalui penelitian ini, diperoleh bahwa seluruh variabel pada data set memiliki pengaruh dalam prediksi model. Diperoleh juga bahwa penalti LASSO pada model Neural Network memberikan peningkatan F1-score menjadi 0,762 dan memberikan nilai AUC yang sebanding dengan model dasar sebesar 0,858. Pada model Random Forest pembobotan pohon penalti LASSO memangkas 1.233 pohon dan diperoleh 267 pohon aktif, diperoleh nilai F1-Score yang sebanding dengan model dasar sebesar 0,734 dan peningkatan nilai AUC menjadi 0,865. Melalui penelitian ini diperoleh bahwa penalti LASSO dapat memberikan performa model yang sebanding atau bahkan lebih baik dari model dasar serta efisiensi pada model. Melalui penelitian ini diharapkan dapat memberikan metode alternatif bagi perusahaan asuransi dalam memprediksi lapse serta menghasilkan model yang informatif.
================================================================================================================================
Lapse is the cancellation of a life insurance policy before its maturity date, which can significantly impact the profitability of insurance companies. Policyholder lapse behavior is influenced by a variety of factors. This study aims to compare the application of Least Absolute Shrinkage and Selection Operator (LASSO) on Random Forest and Neural Network algorithms in predicting life insurance policy lapse. The data used consists of policy and policyholder characteristic data from a life insurance company in the United States for the period January 2012 to November 2017, with a total of 16,140 observations. In the Neural Network, LASSO functions as a regularization technique to reduce weights, perform neuron pruning, and feature selection simultaneously, whereas in Random Forest, LASSO is used to replace the voting mechanism by assigning weights to each decision tree. Through this study, it was found that all variables in the dataset have an influence on model prediction. It was also found that the LASSO penalty in the Neural Network model improved the F1-Score to 0.762 and produced an AUC value comparable to the baseline model at 0.858. In the Random Forest model, the LASSO tree weighting pruned 1,233 trees, resulting in 267 active trees, with an F1-Score comparable to the baseline model at 0.734 and an improved AUC of 0.865. This study concludes that the LASSO penalty can provide model performance that is comparable to or even better than the baseline model while also improving model efficiency. It is hoped that this study can provide an alternative method for insurance companies in predicting policy lapse and producing a more informative model.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Asuransi Jiwa, Feature Selection, Lapse, LASSO, Neural Network, Random Forest, SHAP. ============================================================== Feature Selection, Lapse, Life Insurance, LASSO, Neural Network, Random Forest, SHAP.
Subjects:	Q Science > QA Mathematics > QA401 Mathematical models. Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Actuaria > 94203-(S1) Undergraduate Thesis
Depositing User:	Agnes Putri Victoria Siregar
Date Deposited:	23 Jul 2026 01:06
Last Modified:	23 Jul 2026 01:06
URI:	http://repository.its.ac.id/id/eprint/135394

Actions (login required)

View Item