Human-Centric Machine Learning for Predicting Mental Health Risk in Remote Work Environments: An Industry 5.0 Perspective

Abstract

Remote Work has witnessed a dramatic increase among employees creating numerous advantages yet also causing significant psychological impacts to employees and their mental health. The research will provide a framework for predictive machine learning based on the principles of Human-Centric, Resilience, and Sustainability Industry 5. This research seeks to establish multiple models by examining an organizational survey dataset using logistic regression, random forests, ANN and DNN. All models will be evaluated using 5-fold cross-validation. As such, the results show that, compared to linear models, nonlinear models outperform them, with the Deep Neural Network showing the best discrimination performance, as indicated by ROC-AUC. Feature importance analyses show that the most important predictors are work-life balance, social isolation, sleep quality, workload intensity and organisational support, while demographic variables have a much lower effect. The results of the expected risk scores also show heterogeneity across the workforce, with higher vulnerability among mid-tenure employees and those experiencing multiple psychosocial stressors. Overall, this research shows that the risk of losing one's mental health while working remotely is determined more by human-focused and organisational factors than simply technical considerations.