HIV/AIDS predictive model using random forest based on socio-demographical, biological and behavioral data

Since the beginning of the AIDS epidemic, the HIV virus has infected millions of people, and the count is on the rise with every passing day. This epidemic has affected not only the children and adults but also infants borntoHIV positive mothers. Unfortunately, there is no cure for HIV/AIDS; however, early and accurate prediction methods are required for early treatment and would be helpful to decrease the spread of the disease. Previous studies have investigated the early prediction of HIV infection, mainly utilizing HIV positive patients' medical records. However, besides medical records, other characteristics, such as drug injection, sexual behavior, and other behavioral and biological factors, are also essential to consider while predicting the future acquisition of HIV. In this study, we developed and validated a prediction model for the probability of future HIV acquisition in high-risk groups. We developed prediction models using electronic medical records of patients from Nai Zindagi Trust (NZ), Pakistan. We analyzed the data for socio-demographic, behavioral (drug injection, sexual behavior), and biological features to predict the patients who are at high risk of acquiring HIV. State of the art data mining and machine learning techniques have been applied along with feature selection techniques for HIV risk prediction. The proposed model predicts HIV status with 82% accuracy having an improvement of 10–15% over dominant classifiers, i.e., SVM, Neural Network, J48, and PART. The new risk scores of HIV acquisition could assist health providers in counseling high-risk groups and in targeting intensified testing, treatment, and prevention to people at the greatest risk for HIV infection.