Machine Learning Algorithms for Prediction of COVID-19 in Early Stages Using Explainable AI Approach

The COVID-19 pandemic, in late 2019, profoundly impacted global health, necessitating early detection methods for healthcare professionals. This study analyzes the Mexican Patients’ Dataset (Covid109MPD) to discern optimal features for predicting outcomes among Mexican patients. Different machine learning algorithms like Naïve Bayes, support vector machine, multilayer perceptron, and K-nearest neighbor are evaluated, with subsequent enhancement through feature selection techniques, i.e., info gain and gain ratio. Results highlight the superior performance of Naive Bayes and multilayer perceptron classifiers, achieving 94% accuracy initially without feature selection and 95% with feature selection. These findings offer valuable insights for healthcare practitioners in early COVID-19 diagnosis. Explainable AI, facilitated by SHapley Additive explanations (SHAP) analysis, elucidates feature significance, aiding the interpretability of predictive models.