A Robust Random Forest Prediction Model for Mother-to-Child HIV Transmission Based on Individual Medical History

Abstract

Human Immunodeficiency Virus (HIV) continues to be a leading cause of mortality and reduces manpower throughout the world. HIV transmission from mother to child is still a global challenge in health research. According to UNAIDS, in every 7 girls, 6 are found to be newly infected among adolescents whereby 15-24 years are likely to be living with HIV which is the maternal age and likely to transfer to the child. Machine learning methods have been used to predict HIV/AIDS transmission from mother to child but left behind some important considerations including the use of patient-level information and techniques in balancing the dataset which may impact models’ performance. A robust prediction model for mother-to-child HIV/AIDS transmission is vital to alleviate HIV/AIDS detrimental effects. The Random Forest Machine Learning method was employed based on features from the individual medical history of HIV-positive mothers. A total of 680 balanced data tuples were used for model development using the ratio of 75:25 for training and testing the dataset. The Random Forest model outperformed the most commonly used learning algorithms achieving the performance of 99% accuracy, recall and F1-score of 0.99 and an error of 0.01, thus improving the prediction rate.