During the COVID-19 pandemic, front-line personnel worldwide had tremendous psychological stress compared with the general population. High mental stress may lead to job burnout. This paper starts with gathering a job burnout dataset from medical staff and police officers working in Kuwait during the COVID-19 pandemic using a webbased Arabic version of the Maslach Burnout Inventory questionnaire. The gathered dataset shows that there an elevated burnout rates among the front-line personnel dealing with COVID-19 patients. It utilizes machine learning techniques including AnDE Bayesian, JChaid* decision tree, SVM margin-based, ForestPA Decision forest, and DMLP to predict the presence of job burnout. Then, we present efficient feature subset selection approaches using several metaheuristic methods such as Bat, Cuckoo, PSO, GWO, and CGWO to select the most competent features. Experiments showed that reducing the number of features allows for a better understanding of the underlying model used to make predictions. Results also support the adaptation of deep learning architectures in social sciences when data is relatively small. These results may considerably help to screen out the front-line workers at high risk for job burnout. Keywords��� Job Burnout; COVID-19; Chaotic Grey Wolf Optimization; Machine Learning; Data Mining.