An Automated Identification of Mental Workload in a Person- Independent Setup
Abstract - In the current study, we report recent advances in automated cognitive load recognition based on peripheral physiological signals in person-independent and person-specific modeling scenarios. This research evaluates the applicability of machine learning methods to detect cognitive load induced by three interactive cognitive tasks of different complexity, such as the Stroop test and sets of logical and math problems. We consider an experimental setup to evaluate the proposed method, which builds on the publicly available CLAS dataset. In the experimental evaluation, we assessed the applicability of a range of purposely designed features, with and without feature vector normalization, when used with various classification methods. We report a recognition accuracy of up to 86.2% in the person-independent scenario and up to 87.2% in the person-specific scenario.The best classification accuracy was obtained, in both cases, with z-normalized feature vectors and a Random Forest classifier.
Keywords - Cognitive Load, Physiological Signals, Clas Dataset, Classification, Z-Norm