Loading...
Machine learning models for unobtrusive monitoring of perceived control and the prediction of stress
Citations
Altmetric:
Date
2025-12
Abstract
Perceived control refers to the belief in one's ability to influence outcomes. This belief is closely linked to mental health, as individuals with higher levels of perceived control experience lower levels of anxiety, depression, stress, and general psychological distress. They also tend to perform better in their fields of endeavour than those with diminished levels of perceived control.
This study explores data generated from the use of a mobile application that helps users to assess their perceived control. This application was designed to investigate the implementation of perceptions of control outside a lab-setting. The data is then used to create machine learning models for identifying individuals displaying symptoms of anxiety, depression, stress, and general psychological distress.
Data was collected at the University of Limerick in 2017 and at the University of Ghana from 2023 to 2024, where 106 and 118 participants were recruited, respectively. Participants completed trials and judgements, that enabled them to estimate their perceived control through self-reported ratings and received feedback and reminders that could potentially influence the outcomes in the process.
The models developed included Random Forest, Extreme Gradient Boosting, Gradient Boosting, Decision Tree, k-Nearest Neighbors, and Support Vector Machine. These models were evaluated using a 6-fold cross-validation with hyper-parameter tuning. The main features considered in this study were the numerical values reported for the internal and external perceived control ratings, and socio-demographic variables such as age and gender, as well as experimental variables. The experimental variables described the characteristics of the messages that users received and the deactivation of button clicks in either or both the reminder and feedback messages.
The findings confirm the fact that machine learning models can predict symptoms of depression, anxiety, stress, and general psychological distress based on the perceived control data. The analysis also revealed that there was no added advantage in using the combined DASS-21 score over the individual subscale scores as the basis of labelling the dataset.
Supervisor
Description
Peer-reviewed
Publisher
University of Limerick
Citation
Collections
Files
Loading...
Azaglo_2025_Machine.pdf
Adobe PDF, 6.32 MB
ULRR Identifiers
Funding code
Funding Information
Sustainable Development Goals
External Link
License
Attribution-NonCommercial-ShareAlike 4.0 International