Enhanced Dyslexia Detection Technique Based on TransformerMLP and Modified Animated OAT Optimization Algorithm

Safar M. Alghamdi¹Nermine Mahmoud²Ibrahim Nafisah³Abdelghani Dahou⁴Mohamed Abd Elaziz⁵Mohammad Ghatasheh⁶Ibrahim A. Fares^5*

• ¹Department of Mathematics and Statistics, College of Science, Taif University, P.O. Box 11099, Taif 21944, Saudi Arabia, Taif, Saudi Arabia
• ²Galala University, Suez, Egypt
• ³King Saud University, Riyadh, Saudi Arabia
• ⁴Zhejiang Normal University, Jinhua, China
• ⁵Faculty of Science, Zagazig University, Zagazig, Egypt
• ⁶Middle East University, Amman, Jordan

This paper proposes a dyslexia detection technique that integrates a deep learning model with a feature selection algorithm. To enable multimodal integration, we developed TransformerMLP for hierarchical representation learning from tabular eye-tracking metrics alongside SwinV2 for spatial-temporal feature extraction from fixation maps, creating a comprehensive feature extraction framework. In addition, we enhance the performance of the Animated Oat Optimization Algorithm based on double attractors. To evaluate the performance of the developed dyslexia detection model, a set of experiments is conducted using variant datasets. In addition, we compare the results of the developed model with state-of-the-art methods based on performance measures. The results of the developed model show the excellent capability in detecting dyslexia, with an average accuracy of nearly 92.54% overall tested datasets.