Identifying dyslexia in school pupils from eye movement and demographic data using artificial intelligence

Area of Science:

• Neuroscience
• Computer Science
• Developmental Psychology

Background:

• Previous dyslexia detection methods often rely on limited datasets.
• Existing datasets for dyslexia research lack multi-source integration and sufficient granularity.

Purpose of the Study:

• To introduce a novel, large-scale, multi-source dataset for dyslexia identification.
• To develop and evaluate an artificial intelligence (AI)-based solution for dyslexia detection in primary school pupils.
• To investigate the psycholinguistic significance of various features in dyslexia identification using AI.

Main Methods:

• Collected and annotated a new, largest-ever eye-movement-during-reading dataset, combined with demographic and non-verbal intelligence data.
• Formulated dyslexia prediction as both classification and regression tasks, evaluating 12 classification and 8 regression models.
• Utilized Bayesian optimization for hyperparameter tuning and performed stratified ten-fold cross-validation.

Main Results:

• The combined multi-source dataset significantly outperformed individual data sources for reliable dyslexia detection.
• Multi-layer perceptron, random forest, gradient boosting, and k-nearest neighbor models demonstrated the most acceptable performance.
• Key predictive features identified include IQ, gender, age, and y-axis eye fixation patterns.

Conclusions:

• A novel, comprehensive dataset and AI approach offer a robust solution for dyslexia detection.
• Integrating diverse data sources (eye movement, demographics) is crucial for accurate dyslexia identification.
• Feature importance analysis provides insights into the psycholinguistic underpinnings of dyslexia.