Performance Evaluation of Machine Learning Algorithms for AIDS-Infected Patient Classification

Background: According to UNAIDS (2023), approximately 39.9 million people are living with HIV worldwide, with 1.3 million new cases and 630,000 AIDS-related deaths in 2023. This indicates that HIV/AIDS remains a serious global health threat. Machine learning methods have the potential to improve the accuracy of AIDS infection classification.Objective: This research is aimed to determine the best classification method based on prediction accuracy and to identify the method with the best performance for further analysis.Methods: This research used a quantitative approach by evaluating the performance of machine learning algorithms: Decision Tree, Random Forest, XGBoost, Naive Bayes, and Logistic Regression. Secondary data were obtained from the UCI Machine Learning Repository, comprising 2,000 observations of AIDS patients and 23 variables. Model evaluation used a confusion matrix to calculate accuracy, precision, recall, and F1-score. The best model, logistic regression, was further analyzed with parameter significance tests, odds ratios, and goodness of fit.Results: Logistic regression yielded an accuracy of 88.4%, precision and recall of 90%, and the highest F1-score. Variables significant to AIDS were: time, preanti, symptom, offtrt, and cd420. The model passed the Hosmer and Lemeshow test (p-value = 0.365) with a Nagelkerke R-Square of 0.642.Conclusion: Machine learning approaches, particularly logistic regression, support early detection of AIDS and data-driven medical decision-making.