Predicting Pandemic Fatality Based on Supervised Machine Learning Methods

Jamil AlShaqsi1,Email

Mohamed Borghan2

Osama Drogham3

Gholam Hossein Roshani4,Email

Salim Al Whahaibi5

1Department of Information Systems, Sultan Qaboos University, P.O. Box 20, PC 123, Muscat, Oman.
2Department of Pre-medicine, National University of Science and Technology, P.O. Box 620, PC 130, Muscat, Oman.
3Department of Information and Communication Technology Systems, Al-Balqa Applied University, Al-Salt 19110, Jordan.
4Electrical Engineering Department, Kermanshah University of Technology, Kermanshah 6715685420, Iran.
5Falha Medical Solutions, Muscat, Oman.

Abstract

This study aimed to conduct intensive experiments on predictive machine learning models to predict the parameters behind death due to the coronavirus (COVID-19). Datasets were obtained from Mexican Government and Various machine learning algorithms were used to detect the hidden patterns from the obtained datasets. Feature selection technique was used to optimize the precision of the supervised machine learning algorithms. Given a large employed sample (N = 67300 patients), the 10-fold Cross-Validation technique was used to validate the experimental results. Several metrics were used to measure the algorithms' accuracy and then compare the outcomes of the conducted experiments. Based on the analysis of multiple algorithms, it has been found that J48 algorithm outperformed other algorithms' classification performance. More importantly, it has been confirmed that some parameters significantly contribute to the death of infected patients. Oxygen saturation, pneumonia, and age are the leading predictors for predicting mortality. The obtained results will help in minimizing the risk of death by re-structuring the treatment protocol.