Automatic cognitive load evaluation using writing features: An exploratory study

Abstract

Evaluating cognitive load is a key step in designing adaptive multimedia learning environments. However, there is still a lack of evaluation methods which can not only unobtrusively collect user data without supplement equipment but also objectively, quantitatively and in real time evaluate user cognitive load based on the data. This paper presents a new approach to evaluating cognitive load by combining writing features from free text and machine learning techniques. Specifically, changes in writing features are first investigated across three levels of cognitive load and the results offer some first insights for the potential of writing features to indicate cognitive load changes; further, a single feature is examined to detect which features are most predictive of cognitive load changes, and back-propagation neural networks, along with two feature selection methods, are trained to classify three cognitive load levels with 76.27% accuracy. These results show that writing features are useful for evaluating cognitive load when suitable classifiers are adopted. Relevance to industryThis study provides evidence that cognitive load can affect handwriting features, and also develops an automatic classification method of discriminating different levels of cognitive load on handwriting features using machine learning technologies. The findings can extend the use of handwritten devices and provide a new method for usability testing.