Predicting choice behaviour in economic games using gaze data encoded as scanpath images.

Area of Science:

• Behavioral Economics
• Cognitive Science
• Machine Learning

Background:

• Eye movement data is crucial for understanding decision-making in economic games.
• Traditional analysis of gaze data in these games often uses logistic regression.
• Advancements in machine learning offer new possibilities for analyzing complex behavioral data.

Purpose of the Study:

• To evaluate the effectiveness of deep learning and support vector machine models in predicting decision strategies from eye movement data.
• To develop a method for creating scanpath images that capture gaze dynamics for machine learning prediction.
• To compare the accuracy of these machine learning models against traditional logistic regression.

Main Methods:

• Generating scanpath images from eye movement data to represent gaze behavior dynamics.
• Applying deep learning (DL) and support vector machine (SVM) classification algorithms.
• Utilizing a baseline logistic regression (LR) model for comparative analysis.

Main Results:

• DL and SVM models accurately identified participants' decision strategies before action.
• The proposed approach achieved an 18% higher classification accuracy compared to the baseline logistic regression model.
• Scanpath image generation effectively captured gaze dynamics for predictive modeling.

Conclusions:

• Machine learning, particularly DL and SVMs, offers a significant improvement over traditional methods for analyzing eye-tracking data in economic games.
• Eye-tracking data, when processed with advanced ML techniques, can create valuable information asymmetries in strategic environments.
• The increasing prevalence of eye-tracking technology in consumer applications (e.g., VR) highlights the future importance of this research.