This study evaluated 5,864 adults receiving HIV treatment at the University of Gondar Hospital from 2018 to 2023 and used 7 machine learning models to predict adverse drug reactions. Researchers used patient records and clinical variables, such as immune system status and treatment details to train and compare the models. Women represented 64.04% of participants and men were 35.06%. The random forest model achieved the strongest performance, with a sensitivity of 1.00 and an AUC of 0.9989, showcasing a near perfect ability to identify true cases of harmful reactions. CD4 count, which measures immune strength, was the most influential predictor across analyses. Additional high risk indicators included being male, being younger, spending longer periods on antiretroviral therapy, and not receiving preventive medications for tuberculosis or bacterial infections. A commonly used regimen, TDF 3TC EFV, was also linked to higher rates of adverse reactions. Association rule mining confirmed that these factors frequently appeared together in patients who experienced side effects. The authors reached the conclusion that these patterns form a reliable early warning structure within routine clinical data.