Enhancing the COVID-19 diagnostic for patients using machine learning and voting classifier approaches

The COVID-19 pandemic highlighted the need for prompt diagnostic techniques in order to control the disease’s spread. This study, which focuses on the innovative use of machine learning (ML) models, intends to improve the efficiency and accuracy of diagnosis. Five supervised machine learning models were employed using a dataset acquired from the University of Texas Medical Branch at Galveston (UTMB) consisting of 1987 patient records with 30 vital measurements. The study utilized the ML models, including Support Vector Machine (SVM), Classification and Regression Tree (CART), Random Forest (RF), Artificial Neural Network (ANN), and Voting Classifier (VC). Using the PCR label and the remaining 29 vital measurements as the features, the models were trained to determine whether the feature combination was associated with COVID-19 infection. Data preprocessing included outlier detection, oversampling for class balance, and multicollinearity analysis through Variance Inflation Factor (VIF) analysis, confirming the data quality. By eliminating outliers and correcting the bias through the data preprocessing steps, a refined dataset was prepared for model training. The voting mechanism combined with a Support Vector Machine (SVM), Random Forest (RF), and Artificial Neural Network (ANN) yielded an accuracy of 0.804 and an Area Under Curve (AUC) of 0.85. Dimensionality reduction and feature contribution analysis were performed to obtain a comprehensive understanding of the model building process. These findings point to the potential application of machine learning models in diagnosing COVID-19.