The variable importance feature provided by ensemble methods can be of great value for health insurance companies in understanding the key characteristics of insured individuals that affect the frequency and severity of insurance claims. This, in turn, can help establish pricing criteria, enabling the development of more accurate pricing models. This paper advocates ensemble methods, specifically Bagging and Random Forests, to determine the relevance of risk-related variables in a pricing model within compulsory health insurance schemes. The study focuses on a Moroccan mutual health insurance company operating in the private sector. Using the specified algorithms, we identified that the insured individual’s age, gender, type of care consumed, and presence of chronic illness are the most important variables influencing severity and frequency models. Evaluating both models using RMSE, Random Forest (RF) achieved the best performance for the severity model, while both algorithms showed comparable results for the frequency model.