Identifying Dynamic Spatial Cognitive Strategy by Eye-Tracking Metrics Using Unsupervised Machine Learning and IRT modeling

• In Event: Innovative Cognitive Approaches to Learning and Assessment

Abstract

This study explores latent cognitive strategies in spatial reasoning tasks using unsupervised machine learning on eye-tracking metrics across fixation, saccade, pupil dilation, blink, and response time. The Explainable Deep Clustering Model (EDCM) combines autoencoder and K-Means clustering identified distinct cognitive behavior patterns. To examine strategy use under varying task difficulty, we utilized Item Response Theory (IRT) to estimate latent abilities and item difficulties. Four cognitive strategies emerged: Impulsive Scanner, Effortful Engager, Hyperactive Explorer, and Efficient Processor. Findings show that strategy effectiveness is shaped by the interaction between cognitive strategy, task difficulty, and participant ability, highlighting the importance of strategy flexibility.

Authors

• Jujia Li, University of Alabama

• Wei Huang, University of Alabama

• Kaiwen Man, University of Alabama

• Joni M. Lakin, University of Alabama