EEG-Based Detection of Cognitive Task Engagement During Motor Dual-Task Testing

Dual-task balance assessments, combining motor tasks such as the Timed Up and Go or double-leg stance with a concurrent cognitive task, are widely used to evaluate cognitive-motor interaction. When vocalization of the cognitive task is not feasible or introduces bias, no objective method currently exists to verify whether the subject is genuinely performing the secondary task, potentially compromising assessment validity. This thesis proposes an EEG-based machine learning framework to address this gap. EEG signals were acquired from 13 healthy young adults using the EMOTIV EPOC X wireless headset during both protocols under single-task and dual-task conditions with serial subtraction. After preprocessing (spectral filtering, NLMS adaptive filtering, ASR, ICA with ICLabel), spectral, connectivity, entropy, and time-domain features were extracted and fed to binary classifiers (SVM, LDA, LR, Random Forest) evaluated via Leave-One-Subject-Out cross-validation. Spectral features at frontal and prefrontal electrodes consistently yielded the highest classification performance across both protocols, demonstrating the feasibility of EEG-based cognitive compliance verification during dual-task balance assessments.