A Customized CNN-Based Framework for Learning Disability Detection Using Handwriting Image Classification

Abstract

Handwriting analysis has been widely explored as a supportive computational approach for understanding writing patterns frequently associated with learning difficulties in children. While deep learning techniques, particularly transfer learning algorithms, have demonstrated strong performance in image-based classification tasks, their reliance on large pre-trained models often limits adaptability and computational efficiency. To address these limitations, this study proposes a custom Convolutional Neural Network (CNN) architecture specifically designed for the classification of children’s handwriting patterns from images. The model consists of five Conv2D layers to extract and learn significant spatial features, promoting high classification accuracy while maintaining simplicity and reduced computational complexity. A carefully curated dataset of handwriting samples classified as “correct,” “casual,” and “reverse” was used to evaluate the model. This work does not claim to provide a clinical diagnosis of learning disabilities; rather, it presents a handwriting pattern classification framework that may support educational analysis and assistive screening when combined with validated assessment procedures.