Main Features for Detecting Mental Fatigue in Elderly Gait: A Machine Learning Approach

Background: Mental fatigue is a significant risk factor for falls in the elderly. Rapid, inexpensive, and objective diagnosis of this condition plays a crucial role in fall prevention. Objectives: This study aims to identify the most important gait components for diagnosing mental fatigue in the elderly using a machine learning-based model. The model was validated against a reference standard combining the Stroop test and a Visual Analog Scale (VAS) score ≥ 70, a well-established method for reliably inducing and measuring cognitive fatigue in older adults. Methods: Thirty community-dwelling older adults (19M/11F; age 67.0 ± 4.3 years) were recruited based on strict inclusion criteria: Age 60 - 75 years, at least one fall in the past year, normal cognitive function [Mini-Mental State Examination (MMSE) ≥ 24], and independent ambulation. Participants with neurological or orthopedic conditions were excluded. The sample size was determined based on precedent in comparable machine learning studies involving gait and mental fatigue, with additional robustness ensured through data augmentation techniques. Fifty-six spatiotemporal gait features were extracted before and after standardized mental fatigue induction using the Stroop test with VAS confirmation. A total of 27,720 ternary feature combinations were evaluated using an Optimal Decision Tree model. Results: Analysis of five consecutive gait cycles revealed that a ternary combination of average stride time, minimum stride length, and minimum stance phase percentage could predict the presence or absence of mental fatigue with an accuracy of 92.2% (95% CI: 86.7% - 97.8%). Unlike traditional approaches [principal component analysis (PCA), t-test, and forward-backward selection], the proposed method preserves original features, accounts for interactions, and achieves superior performance on small datasets, making it a more reliable and accurate tool for diagnosing mental fatigue in elderly individuals. Conclusions: The high accuracy and minimal input requirements of this model allow for the use of inexpensive tools, such as 2D video cameras, to enable continuous, real-time, and precise assessment of mental fatigue in fall-prone elderly populations. However, the study has limitations, including a small sample size and reliance on treadmill-based data, which may affect generalizability.