Enhancing the Prediction of MERS-CoV Survivability Using Stacking-Based Method

Abstract

Saudi Arabia has recorded the highest Middle East Respiratory Syndrome Coronavirus (MERS-CoV) infections globally. Nearly 2,000 cases have been recorded in Saudi Arabia, with a high mortality rate, since the outbreak in 2012. The source of the disease remains unclear, and it has been evident that MERS-CoV can spread through communicating, directly or indirectly, with humans or animals. In our study, we evaluated different machine learning models that can accurately predict the probability of a patient's recovery from MERS-CoV. The data was from the Saudi Ministry of Health’s website corresponding to the years from 2015 to April 2018. A stacking-based ensemble learning has been built to increase the performance of individual models. In our study, we examined the following individual classifiers: naïve Bayes, support vector machine, logistic regression, k-nearest neighbour, Bayesian networks, J48, and random forest along with the proposed stacking-based model. The results show that, in most cases, simple machine learning techniques perform well when predicting recovery unlike predicting death cases. The proposed stacking-based ensemble learning method has shown improvement in the prediction of death cases while maintaining a good predictive power for recovery cases. The proposed technique, which is an ensemble learning method, performed best with 0.751 balanced accuracy and 0.750 G-Mean. Predicting the survivability of patients can help in decision-making on prevention and recovery of MERS-CoV.