Development and Comparison of Forecasting Models for Gross Written Premiums of Life Insurance Companies in Thailand Using Data Mining Technique

Abstract

Accurately forecasting gross written premiums (GWP) of life insurance companies is vital for strategic planning, risk management, and policy formulation in Thailand’s insurance industry. Traditional statistical models commonly used in prior research often struggle to capture complex non-linear relationships in financial data. To address this limitation, the objective of this study is to develop and compare predictive models for GWP using data mining techniques. This study is among the first to evaluate model performance using a comprehensive monthly dataset covering a 12-year period from January 2012 to December 2023, totaling 144 records. The Cross-Industry Standard Process for Data Mining (CRISP-DM) methodology was adopted as the analytical framework. Four machine learning algorithms were implemented: Decision Tree, Random Forest, Support Vector Machine (SVM), and Neural Network. The empirical results reveal that the Neural Network model outperformed all other techniques, achieving a Mean Squared Error (MSE) of 1,174,870.83, a Root Mean Square Error (RMSE) of 1,083.91, a Mean Absolute Percentage Error (MAPE) of 2.53%, and a coefficient of determination (R²) of 97.60%. These findings confirm the superior predictive accuracy of the Neural Network model. The study contributes to the advancement of premium forecasting by offering a robust, data-driven approach that supports strategic decision-making, enhances risk management practices, and improves the precision of premium calculation systems for life insurance companies in Thailand.