Using machine learning to detect events in eye-tracking data.

Area of Science:

• Ophthalmology
• Computer Science
• Data Analysis

Background:

• Eye movement data analysis is crucial for understanding visual behavior.
• Current event detection methods require manual parameter adjustments based on data quality.
• This limitation hinders objective and reproducible analysis.

Purpose of the Study:

• To develop a fully automated machine learning method for classifying oculomotor events from raw gaze samples.
• To eliminate the need for manual parameter tuning in eye movement event detection.
• To demonstrate the practical utility of this automated approach in applications like biometrics.

Main Methods:

• A random forest machine learning technique was employed for event classification.
• The classifier was trained using existing manually or algorithmically detected events.
• The method classifies raw gaze samples into fixations, saccades, and post-saccadic oscillations (PSOs).

Main Results:

• The machine learning approach achieved fully automated classification of oculomotor events.
• The method demonstrated practical utility in an eye movement-driven biometric application.
• Performance comparable to manual coding was achieved, surpassing current state-of-the-art algorithms.

Conclusions:

• Machine learning techniques offer superior event detection in eye movement data analysis.
• Automated classification removes the need for user-defined parameters, enhancing objectivity.
• This approach can achieve performance levels equivalent to manual coding.