A Machine Learning Approach for Dyslexia Screening in a Minoritized Language Context: The Case of Catalan

Abstract

Learning to read is a fundamental skill for academic success and a key tool for enabling individuals to fully participate in society. However, approximately 10% of children face difficulties in acquiring this skill due to dyslexia, a neurodevelopmental disorder that affects reading and writing acquisition. Developing dyslexia detection methods is particularly challenging in minoritized languages, where the smaller number of speakers makes it difficult to gather the large datasets typically required to train machine learning models. In this work, we present an approach for screening dyslexia in Catalan using a gamified test that combines linguistic exercises with machine learning techniques. To achieve this, we designed the content of a computer game, collected data from 730 children —155 of whom were diagnosed with dyslexia— who played the game, and developed a prediction model using various machine learning classifiers along with targeted feature selection. Our method achieved the highest balanced accuracy when using a Single-Layer Perceptron (SLP) classifier (87.46%) and a linear Support Vector Machine (SVM) classifier (86.67%), both applied to a selected subset of features. These results highlight the potential for cost-effective, online early screening of dyslexia in children who speak minoritized languages, especially in contexts where collecting large datasets is not feasible. The results have been integrated into a freely available online application.1