Comparative Study of CNN Models on the Classification of Dyslexic Handwriting

Developmental Dyslexia, one of the learning disabilities is a topic of scientific interest in a variety of disciplines such as psychology, speech and language therapy, data science, etc. While the reason for Dyslexia and its symptoms are still being researched by psychologists, data science is providing ways to intervene and detect them with the aid of technological advancements. Dyslexia is a neurological condition that impairs reading comprehension and has long-lasting impacts. But timely detection and intervention programs can alleviate its effects to a certain extent. This study aims to classify images of handwritten English characters into three classes namely: normal, corrected, and reversed, where normal class refers to normal handwriting, and corrected or reversed constitutes handwriting of children with Dyslexia. The dataset used for the study is available publicly on Kaggle. The building of an efficient CNN (Convolutional Neural Network) model for classifying dyslexic handwriting is the major emphasis of this work. This is accomplished by comparing several CNN models and evaluating how well they detect Dyslexia on the same dataset. The proposed CNN approach has demonstrated a sizable improvement in reliably classifying dyslexic handwritten images.