Classification of Binary Class HIV/AIDS Test Results Using Ensemble Learning Models

In HIV/AIDS datasets, there is lack of adequate and imbalanced samples includes repetitive and unnecessary features that cause high dimensionality spaces. To address this problem, selections of features are examined. In this research, we propose an ensemble learning method for the classification of a binary class of HIV/AIDS test results. The backward feature selection (BFS) of the wrapper method is used for feature selection. Five established classifiers are used, namely Gradient Boosting (GB), Multilayer Perceptron (MLP), Random Forest (RF), Extra Tree (ET), and K-nearest neighbor (KNN). For preparation and testing of the model, 10-fold cross-validation is applied. Experiments are carried out on the EDHS-HIV/AIDS dataset. Various performance measurements are used to assess the model's performance. The confusion matrix is used to demonstrate whether the samples are labeled correctly or not. Based on performance evaluation parameters, a review of the results of each classifier is provided. Significant performance enhancements are seen in the results when feature selection is considered to be better on the original dataset comparison to the selected features for all classifier outputs.