Smartphone-based microkinematic feature analysis for mental fatigue detection using machine learning

Mental fatigue impairs cognitive performance and productivity, posing risks in domains such as healthcare, education, and workplace safety. There is a growing need for objective, accessible tools to detect fatigue in real time. Leveraging the widespread availability and sensing capabilities of smartphones, this study presents a cross-platform system (iOS and Android) that detects mental fatigue using fine motor skill tests and a self-report questionnaire. Data were collected from 347 sessions across 166 smartphones, where participants completed tasks before and after cognitively demanding activities. From the raw motion and touch data, 60 features were engineered. Feature selection was performed using a wrapper-based method, and six machine learning algorithms were evaluated: Logistic Regression, Support Vector Machine, K-Nearest Neighbors, Decision Tree, Random Forest, and AdaBoost. Models were validated using a nested k-fold cross-validation strategy. The best-performing model achieved a sensitivity of 0.86 by combining self-reported fatigue indicators (e.g., anxiety and effort levels) with microkinematic features related to handwriting and hand tremor. This multi-dimensional approach demonstrates the feasibility of using smartphone-based motion analysis for fatigue classification. The system offers potential applications in remote health monitoring, educational assessment, and occupational safety. Additionally, the publicly available dataset provides a valuable resource for further research on cognitive and motor function assessment using mobile devices.