Recently, competition in modern markets has reached its climax due to the unrelenting pursuit of organizations and institutions to enhance their productivity and quality. This imposes a heavy psychological burden on workers, which almost leads to negative effects like depression. This study presents a detailed examination of the influence of the Synthetic Minority Over-sampling Technique (SMOTE) on the predictive classification of work-related absenteeism raised by employee stress, with a focus on enhancing human resource optimization. To accomplish this task, six supervised machine learning (ML) algorithms were utilized as baseline classifiers: Naïve Bayes (NB), Multilayer Perceptron (MLP), Partial Decision Tree (PART), J48, Random Forest (RF), and Random Tree (RT). Absenteeism at work, a well-known dataset available in the UCI repository, was utilized for the study. This paper aims to implement and assess the performance of selected algorithms, focusing on classification accuracy, precision, Recall, and F-measure as evaluation indicator metrics. As the results show, all algorithms are effective in classifying worker absences; however, RF achieves high evaluation metrics, recording 93.9% accuracy, precision, recall, and F-measure.