GA greedy time-frequency analysis of electrodermal activity for cognitive state monitoring

Cognitive workload (CW) is defined as the required mental effort to perform a task, which is associated withcapacity-limited cognitive system for processing information in the working memory. The development of aframework to not only minimize human errors, but also maximize performance, is an important applicationof CW estimation and management. It can also help to improve the diagnosis and treatment of neurologicalor cognitive disorders, and to facilitate human-computer interaction. In place of non-invasive nature,sensitivity to cognitive states, robustness against intentional conduct, and the possibility of onlineinvestigation, psychophysiological signal analysis have received special attention for CW estimation. Hence,a novel CW estimation method based on matching pursuit (MP) decomposition of electrodermal activity(EDA) and support vector machine classifier has been proposed. The MP, as adaptive and greedy timefrequencydecomposition, has the advantages of reducing cross terms, improving time-frequency resolutionand enhancing biomedical signal analysis performance [۵, ۶]. Applying two dictionaries, including RnIdentand daubechies wavelet (db۵) at level ۵, sparse coefficients have been extracted from the EDA signals. Then,several statistical and nonlinear features (mean, standard deviation, variance, covariance, and Shannonentropy) have been calculated from the MP coefficients. The proposed method evaluated on EDA of ۳۰healthy students performing an arithmetic task has achieved an average accuracy of ۹۶.۸۰% for two workloadlevels. Moreover, the experimental results have indicated that the combination of complementary informationfrom the different extracted features has enhanced the estimation performance.

کلیدواژه ها:

Matching Pursuit ، Cognitive workload ، Support vector machine ، arithmetic task