A novel cepstral-based technique for automatic cognitive load estimation

The development of a reliable algorithm with low computational requirement is a challenge in cognitive load estimation. To address this challenge, this paper presented a view towards the novel application of cepstrum analysis in workload estimation. The proposed method used minimum number of physiological signals. It integrated amplitude and phase information of electrocardiogram (ECG) signals while reducing feature sample size. A set of cepstral-based ECG features, such as dynamic cepstral warping (DCW), differential energy density, and differential statistical features was proposed for designing a new cognitive load estimation system. The features were processed by principal component analysis and support vector machine in order to discriminate different levels of an arithmetic task. The discriminating capability of the method was evaluated using the ECG recording of 22 healthy subjects. The proposed algorithm achieved high average accuracy of 92.27% and 90.34% for the workload levels determined by the variation in the digit numbers and in the number of carry operations, respectively. In the case of combination both variables, the average accuracy of 90.48% was obtained. Furthermore, comparing between complex and real cepstral measures revealed better performance of the complex measures. These findings indicated the role of phase information in the ECG-based cognitive load estimation. Integrating dynamic and static characteristics of the cepstral coefficients in a multivariate approach improved the performance of the system. It performed significantly better than popular ECG features. The proposed approach provided a trade-off between performance and computational complexity of the estimation system. (C) 2017 Published by Elsevier Ltd.