Machine learning-based classification of viewing behavior using a wide range of statistical oculomotor features.

Area of Science:

• Cognitive Science
• Neuroscience
• Computer Science

Background:

• Oculomotor behavior is influenced by cognitive state and task demands.
• Research increasingly uses data-driven methods to classify eye movement patterns.
• Understanding task-switching costs is crucial for cognitive research.

Purpose of the Study:

• To decode cognitive state and task-switching using machine learning on oculomotor data.
• To identify key oculomotor features predictive of cognitive states and task changes.
• To evaluate the feasibility of data-driven classification of oculomotor behavior.

Main Methods:

• Utilized a large dataset from multiple experiments.
• Implemented state-of-the-art machine learning classifiers.
• Extracted a wide range of oculomotor features for analysis.
• Performed feature ranking to identify important predictors.

Main Results:

• Achieved robust classifier models for decoding cognitive state and task-switching.
• Demonstrated decoding performance independent of image statistics.
• Identified distinct sets of important oculomotor predictors for each classification task.
• Provided feature rankings highlighting the predictive power of various oculomotor metrics.

Conclusions:

• Oculomotor features can be effectively used to decode cognitive state and task-switching via machine learning.
• This data-driven approach offers a feasible method for analyzing complex visual attention patterns.
• Feature importance analysis reveals specific oculomotor correlates of cognitive processes.
• The findings have implications for interpreting decoding results and understanding cognitive control.