Evaluating Dysgraphia Level Using Handwriting Dynamics

This study investigates the use of machine learning for assessing dysgraphia severity through handwriting analysis. Handwriting data were collected from eight distinct tasks per-formed by children diagnosed with dysgraphia and matched controls, recorded using a Wacom tablet. Each sample was manually graded by trained professionals to establish severity levels. Features related to spatiotemporal dynamics, kinematics, and pressure were extracted and used to train an XGBoost classifier, with cross-validation ensuring robust evaluation. The results showed task-specific performance differences, with the highest accuracy of 61.89% for sentence-writing and 60.15% for single-letter writing. Tasks involving longer words or pseudowords showed lower performance due to increased motor and cognitive demands. Low inter-rater agreement among human evaluators highlighted the Subjectivity of manual grading, underscoring the potential of automated, objective assessment methods. These findings support the use of machine learning in dysgraphia evaluation while emphasizing the need for improved task designs and feature optimization for broader clinical and educational applications.