Predicting Cognitive Load with Wearable Sensor Signals

Abstract

This research focuses on predicting the affective state, such as a cognitive load of a person performing cognitive tasks. The predictors included the physiological data, demographics, and personality type available in the CogLoad dataset. Specifically, the chosen physiological data included heart rate, intervals between successive heartbeats, galvanic-skin response, and temperature. We experimented with several machine-learning models. Among the classifiers, the LightGBM achieved the best accuracy of 74.41% and F1-score of 77.10% in detecting the cognitive load.