A Systematic Evaluation of Machine Learning Techniques for Dyslexia Detection Using EEG Signals

Abstract

Dyslexia is a learning disability that develops in childhood, leading to difficulty in reading, writing, and spelling, while having average or above average intelligence. The range of impact can be significant if this condition is not diagnosed early. Various methods have been tested to diagnose this condition, including psychometric assessments, facial and eye movement tracking, magnetic resonance imaging, and electroencephalograms. Electroencephalography (EEG) measures the electrical activity of the brain and contributes to understanding cognitive processes. It is particularly useful in recognizing patterns of neural activity that correlate with dyslexia, providing a cost effective and objective means of diagnosis. This paper attempts to review available diagnostic methods and highlight their shortcomings and gaps in research. It also evaluates machine learning and deep learning models for the classification of dyslexia using EEG data collected from a video-based educational paradigm. The models include Support Vector Machine, Random Forest, Logistic Regression, KNN, Decision Trees, AdaBoost, Naïve Bayes, CNN, LSTM, BiLSTM, and GRU. The best models in capturing EEG data patterns were CNN and LSTM, while the best classical baselines were SVM and AdaBoost. The proposed method showed a weighted F1 score of 0.99, proving reliability.