Failing to classify risk accurately can cost insurance companies a fortune, and the health insurance industry is no exception, especially when dealing with insureds with chronic illness or those more likely to incur heavy claims. This paper discusses the classification of policyholders in the Moroccan health insurance context according to their risk levels. To do so, we used Deep Learning and Machine Learning approaches, namely Multilayer Perceptron (MLP), Extreme Gradient Boosting (XGBoost), and Support Vector Machine (SVM). These models were used to categorise policyholders into risk groups. The dataset was sourced from a Moroccan private-sector insurance company and initially contained 96,540 records. A series of preprocessing steps was performed to address data imbalance and improve ML model performance, including undersampling. The models were evaluated using three performance metrics: recall per class, accuracy, and F1-score. The findings indicated that all models produced strong results; however, the MLP neural network achieved the highest overall performance. XGBoost also demonstrated commendable performance, whereas the SVM model showed a lower F1 Score and required considerably more computational time than the other models.