Comparative Performance Analysis of Machine Learning Classifiers on COVID-19 Datasets: Exploring the Effect of Data Imputation, Cross-Validation, and Principal Component Analysis

Handling missing data is a vital aspect of data preprocessing that interests data scientists and machine learning researchers/practitioners. Usually, there is the need to select the best imputation technique to replace the missing values in the dataset to enhance the data quality, which ultimately enhances the results of machine learning tasks. Other activities performed during the training of ML models, such as cross-validation and dimensionality reduction, are also critical to achieve the best results. In this paper, by using three COVID-19 datasets, we explore the effect of applying different imputation methods, cross-validation and principal component analysis (PCA) on the classification performance of three machine learning models. The three machine learning models are Deep Multilayer Perceptron (DMLP), Extreme Gradient Boosting Trees (XGBoost) and Support Vector Machines (SVM). The results show that using the regression imputation technique without PCA and cross-validation significantly increased the performance of these three ML models, especially the DMLP. The results of the experiments offer vital insight for data scientists when using these three ML models for binary classification tasks involving medical datasets.