Machine Learning Based Screening for Psychological Distress Using a Perceived Control Mobile App

Perceived control refers to the belief that individuals can take the necessary actions to achieve desired outcomes in various situations. This belief is closely tied to mental health, as those with a higher sense of perceived control generally experience lower levels of anxiety, stress, boredom, and depression, and they often perform better in their respective fields compared to those with a diminished sense of control. This paper investigates the use of data generated from a mobile application aimed at evaluating users’ perceptions of control with the goal of developing machine learning models that can predict signs of psychological distress among participants. The data for this study was collected during 2023 and 2024 in Ghana with 118 participants. Participants’ levels of perceived control were evaluated through a series of tasks, referred to as trials and judgments, which included self-reported numerical ratings that estimated how much users felt the outcomes were contingent on their actions, as well as the influence of external factors on these outcomes. Participants also received various feedback and reminder messages that suggested additional potential influences on the results. The data collection and subsequent data cleaning resulted in a dataset of 401 valid samples. The models developed were the Random Forest, Extreme Gradient Boosting, Gradient Boosting, Decision Tree, K-Nearest Neighbours, and Support Vector Machine, employing a 6-fold cross-validation method with hyperparameter tuning. The findings underscore that machine learning models can predict symptoms of general psychological distress based on perceived control data. The deeper analysis also revealed that the perceived control reported was affected not only by judgements but by app configurations such as the deactivation of button clicks present in the reminders and feedback during the experiment.