An Adaptive Dyslexia Reading Intervention System Using Hybrid Readability Modeling, Error Classification, and Reinforcement Learning

Abstract

Dyslexia is a type of learning disorder that occurs in individuals due to difficulties in reading fluency and comprehension. It can affect an individual’s learning outcome in an important way. Therefore, early prediction and classification of dyslexia are very important to reduce its impact. In this context, this study suggests a data-driven model for early prediction and classification of dyslexia among individuals based on interaction data.. A hybrid feature vector consisting of 15 linguistic readability metrics and 384-dimensional Sentence BERT embeddings is employed for accurate readability score prediction (RMSE = 0.684, R² = 0.553). K-Means clustering is utilized for grouping reading passages into five different levels of difficulty based on K-Means clustering. Errors related to dyslexia, i.e., substitution, deletion, inversion, and transposition, are classified using orthographic features and a Random Forest classifier, achieving 98.03% accuracy for 1.44 million synthetic errors. Epsilon-Greedy Multi-Armed Bandit is employed for adaptive content delivery based on trends of learner performance. Simulations were conducted for accuracy improvement, and it was observed that the proposed adaptive reading intervention system consistently improves accuracy over a static approach, with an average improvement of +0.1648 over a static approach for five randomized simulation trials. Furthermore, error clustering identifies three cognitive sub-types for developing remediation plans for individuals with dyslexia. This holistic approach for individuals with dyslexia is a stepping stone for developing more accurate and effective intervention plans for individuals with dyslexia. The experimental results show better classification accuracy for learners with different levels of risk. Therefore, this model can be helpful in early prediction and classification of learners with dyslexia.