Early Detection and Risk Assessment of Dysgraphia in Sinhala-Speaking Children

Dysgraphia, a specific learning disability marked by difficulties in handwriting, significantly affects academic performance and self-esteem. Early detection and intervention are critical but often hindered by the lack of diagnostic tools customized for the Sinhala linguistic and cultural context. This research introduces a multi-modal sequence approach to address this gap, leveraging handwriting samples, cognitive evaluations, and linguistic tasks to identify early signs of Dysgraphia. The proposed diagnostic framework integrates two core models: a convolutional neural network (CNN) for handwriting sample analysis and a gradient boosting classifier to evaluate cognitive, behavioral, and personal data, assessing Dysgraphia risk levels. The dataset comprises 373 digitized handwriting samples from 84 primary school children, including 73 dysgraphic and 300 non-dysgraphic samples. These were collected from a local pediatric hospital and primary schools, using psychological assessments from Indian contexts adapted to Sinhala language materials. Handwriting samples were preprocessed by converting them to binary format and applying color inversion for analysis. The Dysgraphia detection model demonstrated an accuracy of 96%, while the Dysgraphia severity assessment model achieved 87%. This multi-modal sequential approach enhances diagnostic precision and reliability, enabling timely intervention and tailored academic support. By incorporating advanced AI techniques, the proposed system addresses a significant need in Sri Lanka’s educational landscape, contributing to more effective learning support for children with Dysgraphia.