Predicting Vehicle Insurance Premiums Using Linear Regression, XGBoost, and Neural Networks : A Comparative Study of Predictive Power

Sammanfattning

As machine learning gains traction in the insurance industry, its potential to improve risk assessment and pricing accuracy continues to grow. This study explores the application of three machine learning models, Linear Regression, XGBoost, and Neural Networks, to predict vehicle insurance premiums using structured data collected from Länsförsäkringar and Transportstyrelsen. The central research question investigates which model offers the most accurate predictions. The methodology involves data preprocessing, feature engineering, and model evaluation using RMSE, R2, and SMAPE metrics. Results show that XGBoost achieved the highest accuracy with the lowest SMAPE3 (7.91%), followed by the Neural Network (8.83%) and Linear Regression (13.23%). These findings highlight that adding more complexity in a model can make it easier to handle structured insurance data and capturing nonlinear relationships. The study underscores the potential of ML to enhance data-driven decision making in insurance pricing and provides a foundation for further exploration in actuarial analytics.