Cervical cancer is a malignancy associated with human immunodeficiency virus, characterized by abnormal cervical cell mutations. Machine learning techniques offer valuable support for early detection and prediction of cervical cancer, potentially lowering screening and treatment costs. This study specifically targeted women living with human immunodeficiency virus, aiming to identify the most significant predictors of cervical cancer and to determine the most effective supervised machine learning model for its prediction within this population. This study employed a multi-center, cross-sectional design using a secondary dataset from the smart care systems of four antiretroviral therapy clinics in central Debre Markos town. To determine the most relevant predictors, seven machine learning models, Logistic Regression, Random Forest, K-Nearest Neighbors, Support Vector Machine, Decision Tree, Extreme Gradient Boosting, and AdaBoost were implemented to identify the top-performing model. Model performance was assessed using the confusion matrix and the Area under the Receiver Operating Characteristic Curve. The findings indicated that adherence at enrollment, screening visit type, and nutritional status, months on anti-retroviral therapy, follow-up status, and weight were highly important predictors of cervical cancer. Among the evaluated models, the K-Nearest Neighbors model outperformed the others, achieving the highest accuracy of 98% and an Area under the Receiver Operating Characteristic Curve of 0.68. As demonstrated in this study, the K-Nearest Neighbors model showed the best performance in effectively predicting cervical cancer among women living with human immunodeficiency virus. Strengthening nutritional support interventions, improving follow-up mechanisms, and enhancing anti-retroviral therapy adherence counseling programs may collectively contribute to reducing the risk of cervical cancer among women living with human immunodeficiency virus. Future research should focus on validating the predictive model across diverse geographic regions and healthcare contexts to enhance its generalizability, robustness, and practical applicability.
Mengistie et al. (Fri,) studied this question.