Grid search in hyperparameter optimization of machine learning models for prediction of HIV/AIDS test results

In this work, we propose hyperparameters optimization using grid search to optimize the parameters of eight existing models and apply the best parameters to predict the outcomes of HIV tests from the Ethiopian Demographic and Health Survey (EDHS), HIV/AIDS dataset. The core challenge of this work is to find the right or optimum parameter values that generate the optimal model and uncertain training computing costs and test predictive models using various values of hyperparameters. To overcome these challenges, we explore the effects of hyperparameters optimizations by applying a proposed grid search hyperparameter optimization (GSHPO) on the considered models to robust the prediction power. An extensive number of experiments are conducted to affirm the feasibility of our proposed methods. These experiments are done in two separate phases. In the first phase, we test our method with the selected models before hyperparameter optimization is applying (using the default parameters). The second phase of the experiment is done after the hyperparameter optimization is applying (using GSHPO). During the experiment, the 10-fold cross validation technique is used to solve the bias of the models. The proposed system helped to tune the hyperparameters using the grid search approach to the prediction algorithms. Several standard metrics are used to assess the method's efficiency, like accuracy, precision, recall, f1-score, AUC-ROC, MAE, RMSE, R2 and confusion matrix to compare results of each experiments. The results obtained by after applying 10-fold cross validation techniques and the proposed GSHPO are promising. Our findings suggest that the hyper-parameters of tuning models have a statistically important positive impact on the models' prediction accuracy.