Optimization of machine learning algorithms hyper-parameters for improving the prediction of patients infected with COVID-19

Abstract

In the modeling domain, the selection of appropriate hyper-parameters for classification or prediction algorithms is a difficult task, which has an impact on generalization capacity and classifier performance. In this paper, we compared the performance of five Machine Learning (ML) algorithms from different categories namely: SVM, AdaBoost, Random Forest, XGBoost and Decision Tree. In the first experiment, we adopt a default setting of each model for training and testing. In the second experiment, we use the GridSearch function to find an optimal configuration of the model. The experiments are performed on dataset of anonymous patients with or without COVID-19 disease. The used dataset is obtained from the Albert Einstein Hospital in Sao Paulo, Brazil. To evaluate the reached results, we used different performance evaluation metrics such as: accuracy, precision, recall, AUC and F1-score. The results of the proposed approach have shown that the optimization of the hyper-parameters of the studied learning models leads to an improvement of 18% in terms of Recall.