Prediction of Workload under Distraction using Supervised Learning Algorithms

The complexity of activities in high-risk affairs is increasing daily because of advances in pervasive computing and contemporary interactive technologies. Numerical dexterity and accuracy are critical in various fields such as medical emergency, flights and space travel, radiation sensitive substance managing, defence and border security. Previous research has emphasised the role of attention involving number crunching and the mental exertion it entails. And yet, there is a dearth of research quantifying cognitive load while manipulating stimuli according to the proximity and similarity principles of Gestalt theory. The present research attempts to bridge this gap by capturing neurophysiological data during numerical estimation tasks with induced distraction by conflicting Gestalt perception and then employing predictive analytics methods. Pupillometry and Electroencephalography (EEG) are neurophysiological tools extensively employed in the study of mental load in various settings. This study reviews two dual segregation algorithms like eXtreme gradient boosting (XGBoost) and AutoML that are used for cognitive load identification. The dataset considered the power spectral density (PSD), motivation index (Frontal Alpha Asymmetry), relative power of EEG frequency bands, and pupil dilation. It was seen that as cognitive task load increased the Low beta frequency EEG waves (12.5-18 Hz) became more prominent, and were most active in the frontal regions. Right pupil dilation and frontal theta PSD were also noticed along with that. The significance of the study is to select the fewest or most informative modalities for monitoring in actual working contexts.