Abstract. The application of machine learning models for the analysis of structured survey data is one of the key directions in the development of data-driven analytical systems. However, despite the high predictive performance of many classification algorithms, the lack of interpretability of their decision-making processes limits their practical applicability. Therefore, the joint evaluation of model performance and interpretability is an important research problem. In this study, a comparative analysis of classical machine learning classifiers was conducted using structured survey data. The models considered include Logistic Regression, k-Nearest Neighbors (kNN), Decision Tree, and Gaussian Naive Bayes. The models were trained on stratified training and test datasets, and their performance was evaluated using Accuracy, Precision, Recall, F1-score, ROC-AUC, and LogLoss metrics. In addition to predictive performance, the decision-making logic of the models was analyzed. For this purpose, the explainable artificial intelligence method SHAP (Shapley Additive Explanations) was applied. The results of global and local interpretations showed that model predictions are driven by the combined influence of several key features. Experimental results demonstrated that Logistic Regression provides the most balanced and reliable performance in terms of both predictive accuracy and probabilistic calibration. Furthermore, the explainability analysis confirmed that model decisions are based on meaningful data-driven factors. The proposed approach can be applied in the development of analytical and decision support systems based on structured data.

Keywords: machine learning, classification, explainable artificial intelligence, structured survey data, logistic regression, decision tree, kNN, Naive Bayes, SHAP.