This study proposes a multi-criteria comparison of the main supervised classification algorithms in order to evaluate their performance according to various aspects, including accuracy, robustness, computational complexity and interpretability. The analysis highlights that Random Forest and SVM offer high accuracy but remain computationally expensive, while Naïve Bayes and Decision Tree prioritize speed and transparency. As no model is universally optimal, the choice depends on the application context and specific constraints. The integration of explainability and automatic optimization tools, notably via AutoML and hybrid models, opens up promising prospects for improving the selection and adaptation of algorithms to real needs.