Explainable AI Model for Medical Insurance Premium Prediction using Ensemble Learning with SHAP Analysis
Aditya Bhongade, Yogita K. Dubey, Prachi Palsodkar, Punit Ratnakar Fulzele · 2024
Medical insurance premium prediction is a crucial task for both insurers and consumers. This paper presents ensemble-based techniques to predict medical insurance prices effectively. The trained models were evaluated based on various quantitative metrics. Explainable AI using SHAP analysis is carried out for model interpretation. Histogram Gradient Boosting produced training and testing accuracies of 91.06% and 85.28%, respectively. The Random Forest (RF) model achieved training and testing accuracy values of 96.48% and 81.65%, respectively, while the Gradient Boosting (GB) model resulted in training and testing accuracies of 87.68% and 83.84%. Histogram Gradient Boosting (HGB) performed best among the techniques in terms of error metrics, producing mean squared error and root mean squared error measures of 6454819.22 and 2540.63, respectively, which are the lowest among the techniques used. On the other hand, Random Forest (RF) produced the highest mean squared error and root mean squared error values of 8045175.00 and 2836.40, respectively, but had the lowest mean absolute error of 1210.14.