Galvanic Skin Response signal based Cognitive Load classification using Machine Learning classifier

Abstract

Cognitive load (CL) classification is an important research issue in the human-computer interaction paradigm. It is evident from recent research that Galvanic Skin Response (GSR) can be used to sense cognitive load. CL analysis is important for understanding the mental growth of the child and the psychology of patients going through the different traumatic situation. Inspired by such a novel application, this model is designed. In this work, a technique has been demonstrated to measure and evaluate the level of human cognitive load for different tasks by collecting GSR from 40 student participants. The students are asked to sit for a test to solve three different tasks — reading comprehension, solving mathematics, and cracking Sudoku. Time-domain features have been extracted from participant’s GSR signals while these tasks are being analyzed. Some parameters such as Correlation Dimension (CD), Lempel-Ziv Complexity (LZC), Hurst Exponent (HE) and Shannon Entropy (SE) are used to analyze CL and these are used as features while accomplishing classification. The Level of stress or cognitive load is strongly observed with the one-way ANOVA test and box-whisker plots. Next, several machine learning algorithms are used to classify the various level of cognitive load. Using the 10 fold cross-validation and Naive Bayes algorithm 91.5% accuracy is obtained.