A Machine Learning Approach to Classify Driver Mental Workload as Assessed by Electroencephalography through Infrared Thermal Imaging

Mental workload (MW) describes the brain commitment needed to accomplish a task. The ecological monitoring of MW allows to potentially support individuals in several ergonomic applications, such as Advanced driver assistance systems (ADAS). In this field, the employment of noninvasive contactless techniques to measure the MW is advantageous to not restrain the driver. To this aim, Infrared Thermal Imaging (IRI) is highly suited thanks to its contactless and lightweight features. This study aimed at estimating drivers’ MW through a machine learning (ML) framework based on facial IRI features. Electroencephalography (EEG), a noninvasive technique that measures the neuronal electrical activity related to brain activations, was used as gold standard to assess the MW. Particularly, the power of the β frequency band was considered as indicative of the MW level. A Support Vector Machine (SVM) classifier with linear kernel was used to classify two levels of MW (i.e., High MW vs. Low MW) from facial IRI features. To test the generalization performances of the method, the leave-one-subject-out cross-validation was performed. The method delivered a cross-validated accuracy of 73.6%, with a sensitivity of 68.0% and a specificity of 79.2%. This method could be effectively employed to improve the driver’s ergonomics for ADAS applications, fostering the traffic accidents preventions.