Comparison of Machine Learning Methods to Detect Stress and Participation for Children with Different Special Needs During Serious Game-Based Therapy: A Observational Study

Abstract

Objective: Serious games have shown promise as therapeutic tools for children with different special needs. However,
understanding how children feel and participate in a game is also important. This study used machine learning
(ML) methods to classify stress in children with different special needs and their participation in serious game-based
therapies.

Methods: This cross-sectional observational study was conducted at the Pediatric Rehabilitation Laboratory of the
Department of Occupational Therapy between March and May 2023. Physiological signals such as blood volume pulse,
electrodermal activity, and skin temperature were collected from 25 children with obstetric brachial plexus injury, dyslexia, intellectual disabilities, and typically developing children during game therapy. Using these physiological signals,
12 ML models were applied to classify children’s stress and participation. Descriptive statistics (mean, SD, frequencies)
were used to summarize participant characteristics. Model performance was evaluated using metrics such as accuracy,
precision, recall, and F1 score.

Results: The results demonstrate that the k-nearest neighbor (KNN) classifier after an autoencoder resulted in the highest F1 scores of 66% and 63% for stress and participation classification, respectively. Furthermore, the eXtreme Gradient
Boosting (XGBoost) model achieved the highest F1 scores of 91% and 86% for the no-stress and no-participation classifications, respectively. When both minority and majority classes were taken into consideration, using KNN following
an autoencoder yielded better results with average F1 scores of 68% and 65% for stress and no-stress and participation
and no participation, respectively.

Conclusion: This study shows that ML methods are effective in classifying children’s stress and engagement states using
physiological signals.

Cite this article as: Coskun B, Ay Ş, Erol Barkana D, Uzun İ, Bostancı Seçkin H, Tarakcı D. Comparison of machine learning methods to detect stress and participation
for children with different special needs during serious game-based therapy: A observational study. Arch Health Sci Res. 2025, 12, 0136, doi: 10.5152/
ArcHealthSciRes.2025.24136.