Mental illness considerably affects technology professionals in the United States, impacting one in five adults yearly. This research forecasts the presence of mental illness in this population using supervised machine learning, namely, Logistic Regression, K-Nearest Neighbor, Naïve Bayes, Decision Tree, and Random Forest algorithms. Specific metrics and the Social Determinants of Health theoretical framework are used in assessing the performance of these algorithms and predictors. ‘Sex,’ ‘readiness to seek professional treatment,’ ‘employer-provided mental healthcare,’ and ‘mental health history of employees’ emerged as the strongest predictors in this study. The K-Nearest Neighbor and Random Forest algorithms emerged as the most suitable predictive models. Experimental and control models were developed to facilitate the comparison of results. The experimental models recorded precision values ranging from 46 to 72%, recall values from 40 to 74%, and accuracy scores from 54 to 67%. The control models recorded precision values from 76 to 92%, recall values from 80 to 94%, and accuracy scores from 80 to 90%. Intriguingly, the performance and accuracy of the machine learning models did not significantly increase when a theory-driven approach to feature selection was employed. A paired t-test revealed a statistically significant difference in accuracy scores, challenging conventional assumptions about the role of theory in this context. This study advances the use of machine learning in mental health research and lays the foundation for theory-guided research in this field.