Comparative analysis of machine learning classifiers

This study presents a comparative analysis of five machine learning classification algorithms: support vector machine (SVM), multilayer perceptron (MLP), classification and regression tree (CART), k-nearest neighbors algorithm (K-NN), and naive Bayes classifier (NB) across four datasets from various domains. Using nested cross-validation, the research evaluated classifier performance on Heart Disease, German Credit, Spambase, and Online Shoppers Purchasing Intention datasets. Results demonstrated that no single classifier consistently outperformed others across all datasets and selection should be based on dataset characteristics and application requirements. Dataset characteristics emerged as the primary factor influencing performance, with class imbalance proving particularly problematic. Training efficiency analysis revealed that simpler algorithms can maintain competitive performance with lower computational costs.