Advanced Strategies for Predicting and Managing Auto Insurance Claims using Machine Learning Models

Abstract

The high severity of automobile claims, which continues to rise, necessitates developing novel approaches for effectively handling claims. Machine Learning (ML) represents an essential solution to this issue of concern. As improving customer service remains the primary goal of auto insurers, the companies in question have naturally begun to adopt and use ML to comprehend better and evaluate their dataset more efficiently. This paper contributes scientifically to the pricing of car insurance, in particular, it focuses on the modeling of the total claims amount by ML models such as Support Vector Regression (SVR), Extreme Gradient Boosting (XGBoost) and Multi-Layer Perceptron (MLP). Further, a comparative analysis will help in this case by opting for statistical metrics ( e.g. Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE),...) as well as stochastic ones ( e.g. Difference Score, Weight Difference Score,...) in both train and test datasets. The result shows that the SVR algorithm, originally tuned by Randomized Search CV, achieves excellent precision and surpasses other models tested, as seen in the Taylor diagram. This model, by contrast, shows less efficient visual distribution of predictions than XGBoost and MLP algorithms. The ultimate value of this study resides in the profound analysis of the dataset, which can offer insurers adequate comprehension to manage these losses effectively.