Multilingual AI-Based Dyslexia Detection

Urdu handwriting images were retained without OCR-based correction or linguistic normalization to preserve visually relevant characteristics such as dots, curves, loops, stroke formation, spacing, and character morphology. The deposited collection includes the accepted handwriting images together with a metadata CSV and README documentation. The metadata records image identifiers, original filenames, class labels, image dimensions, camera information where available, file format, file size, quality-control status, preprocessing information, and integrity checksums.

Participant identifiers are intended to remain anonymized. The dataset can support research in computer vision, deep learning, transfer learning, image classification, handwriting analysis, educational AI, low-resource language processing, and bilingual or multi-script dyslexia-screening methods. It may also be used to compare preprocessing, feature-learning, augmentation, validation, and classification strategies.

This dataset is associated with the published research article: Kashif, M., Haider, Z. M., Muneer, I., Kumar, D., Tahir, T., & Shafi, J. (2026). “Optimizing Deep Neural Models for Early Dyslexia Detection Using Novel Bilingual Handwritten Dataset.” Engineering Reports, 8, e70867. doi.org/10.1002/eng2.70867. Because the dataset concerns handwriting collected from children, reuse should comply with the applicable ethical approval, consent conditions, privacy safeguards, and repository access conditions.