A Genetic Algorithm-Based Ensemble Framework for Dyslexia Prediction from Gamified Tests

Md Moinul Islam, Monjurul Haque, Md Rashadur Rahman

Abstract: Dyslexia, a neurodevelopmental disorder characterized by reading and writing difficulties despite normal intelligence, remains underrecognized due to its subtle and diverse presentation. Early identification and targeted intervention are crucial for improving educational and life outcomes, yet existing diagnostic approaches often lack precision and efficiency. To address this gap, we propose an innovative ensemble feature-selection approach (d-GAP) which combines Principal Component Analysis (PCA) with an enhanced Genetic Algorithm (eGA) incorporating an optimized fitness function. This novel methodology effectively reduces dimensionality and identifies the most discriminative features from a benchmark gamified dataset, significantly enhancing predictive performance. Evaluation of multiple machine learning classifiers using d-GAP demonstrates state-of-the-art accuracy of 99.5% on the benchmark dataset with Random Forest, outperforming conventional and recent approaches within this context. Comprehensive comparative analysis across various performance metrics highlights the robustness and superiority of the proposed ensemble method. Our findings provide a highly accurate, efficient, and scalable diagnostic tool that can facilitate early detection, timely intervention, and improved quality of life for individuals with dyslexia. For reproducibility, the source code is available at GitHub.

External IDs:doi:10.1007/s42979-025-04640-8