Contemplate on ECG signal and enhanced automatic cognitive workload estimation using cost-effective and robust method

The growing use of modern devices increases the risk of cognitive overload, so cognitive workload estimation is required for preventive strategies. However, providing a reliable system in the presence of individual differences, real human affecting factors, and considering practical requirements is a big challenge. In state-of-the-art works, either the estimation generalizability is low due to the use of subject-dependent and hand-craft signal analysis manners or the generalizability or computational cost is high due to the use of subject-dependent evaluation and multi-modal signals. Hence, this study proposes two subject-independent models which leverage the capability of convolutional neural network (CNN) and hybrid CNN-long short term memory to capture spatial and temporal information, signal dependencies and the low computational complexity of single Electrocardiogram modality in the presence of mental fatigue. In comparison with previous methods, it has four distinct characteristics: independent from subjects, independent from feature extraction and classification approaches, cost-effective approach due to the use of single lead, and robust against mental fatigue interference as a source of undesirable variability. The capability of the method has been evaluated on 84 healthy subjects performing ten stages of arithmetic task. Furthermore, the effects of different structures and hyper-parameters of deep learning have been evaluated. This method has achieved a high average accuracy rate of 95 % using the hybrid method across a large number of subjects and other interfering factors. The comparative study with other subject-independent and single modality models has demonstrated approximately 40 % improvement and more generalized performance using a higher number of subjects.