Supervised classification algorithms play a crucial role in machine learning, offering solutions to a wide range of problems such as fraud detection, medical diagnosis and recommender systems. This article proposes a multi-criteria evaluation of seven representative algorithms, including traditional methods (Naive Bayes, Decision Tree, Random Forest, KNN, SVM) and modern approaches (Gradient Boosting, MLP). The evaluation is based on twelve key dimensions, such as accuracy, simplicity, scalability, and robustness against noisy data. Based on an in-depth analysis of the existing literature, this study highlights the strengths and limitations of the various algorithms in a variety of contexts. While models such as Random Forest and Gradient Boosting excel in complex environments, simple algorithms such as Naive Bayes and Decision Tree stand out for their speed and ease of use. Specific recommendations are provided to guide the choice of algorithms according to application constraints. In addition, research perspectives are proposed for solving current challenges, such as optimizing scalability, improving interpretability and developing hybrid models integrating the strengths of several approaches.