Predicting Cartographic Symbol Location with Eye-Tracking Data and Machine Learning Approach.

Area of Science:

• Cartography
• Human-Computer Interaction
• Cognitive Science

Background:

• Visual search is crucial for map reading, influenced by map design and human perception.
• Understanding symbol location (central vs. peripheral) is key for effective map use.
• Eye-tracking offers objective data on visual attention during map interaction.

Purpose of the Study:

• To determine if eye-tracking data and machine learning can predict the location of cartographic symbols (central or peripheral).
• To identify which eye movement features are most predictive of visual search focus.
• To explore the potential for gaze-aware cartographic interfaces.

Main Methods:

• Analysis of two eye-tracking datasets from visual search tasks on maps.
• Extraction and standardization of eye movement features (fixation duration, saccade amplitude, gaze dispersion).
• Application of feature selection and machine learning classification algorithms (e.g., Random Forest, Gradient Boosting, AdaBoost).

Main Results:

• Models achieved high accuracy (e.g., 0.822) and ROC-AUC (> 0.86) on the first dataset, with AdaBoost and Gradient Boosting performing best.
• The second dataset presented more classification challenges, though some models showed high recall.
• Fixation standard deviation (vertical axis) emerged as the most predictive feature for gaze dispersion and search focus.

Conclusions:

• Gaze behavior, particularly vertical gaze dispersion, reliably indicates the spatial focus of visual search on maps.
• Machine learning models can effectively predict cartographic symbol location from eye-tracking data.
• Findings support the development of adaptive, gaze-aware cartographic interfaces for improved usability.