رضا حقيقي خرم

The insurance industry, particularly life insurance, plays a fundamental role in ensuring households’ financial security against long-term economic risks, an‎d its sustainability depends heavily on customer retention an‎d the reduction of policy surrender rates. With increasing economic complexity an‎d intensified competition in the insurance market, the use of data-driven methods for customer behavior analysis an‎d churn risk management has become an undeniable necessity. The purpose of this study is to identify the key factors influencing life insurance policy surrender an‎d to segment policyholders based on their level of churn risk. The dataset used in this research consists of information from 433,253 issued life insurance policies during the period 2018–2022, including economic indicators, demographic characteristics, an‎d policy-related features. In the preprocessing stage, the data were cleaned an‎d the problem was formulated as a binary classification task between surrendered an‎d active policies. To address class imbalance, oversampling of the minority class was applied, an‎d three decision-tree-based boosting algorithms were implemented, namely Gradient Boosting, Light Gradient Boosting Machine (LightGBM), an‎d a categorical feature–specific boosting model. Hyperparameter tuning an‎d decision threshold optimization were performed using the Optuna optimization framework, an‎d the optimal model configuration was selec‎ted based on maximizing the F1-score. eva‎luation results on the test dataset indicated that both Gradient Boosting an‎d LightGBM achieved an accuracy of approximately 98%, an F1-score close to 89%, an‎d high values of the area under the ROC curve, demonstrating the strong predictive power of the models. Furthermore, clustering analysis was conducted to divide policyholders into three risk categories (low, medium, an‎d high risk), revealing statistically significant differences in surrender behavior among the groups. The findings suggest that the proposed framework can serve as an effective decision-support tool for churn risk management an‎d customer retention in the insurance industry.

ريزش مشتري , يادگيري ماشين , الگوريتم بهينه سازي , بيمه زندگي

Customer Churn , Machine Learning , Optimization Algorithms , Life Insurance Policies

Reza Haghighi Khorram

Alireza Aliahmadi