A Comparative Analysis of Machine Learning Algorithms for Dyslexia Detection

Abstract

A number of machine learning models are available today to identify disease predictions. But some diseases, if identified through computing, could drastically change the medical industry. One such disease is dyslexia. This study aims to create a machine learning-based predictive model for early dyslexia detection. Utilizing a dataset comprising quiz scores and survey responses related to language skills, memory, speed, visual discrimination and audio discrimination, alongside computed survey scores and dyslexia likelihood labels, five machine learning models were assessed. Models include Decision-Tree, Random-Forest, SVM, Random-Forest with Grid Search, and SVM. with Grid-Search underwent evaluation based on error, precision, and recall metrics. Results revealed that the Random-Forest with GridSearch model demonstrated superior performance. Subsequently, a final model was developed using Random-ForestClassifier with Grid-SearchCV. Testing on a new dataset yielded a 5.8% error rate in dyslexia predictions. Furthermore, an interactive user-friendly interface, Input test, was developed to simplify parameter input and result interpretation. This research advances dyslexia detection methodologies, potentially offering avenues for early intervention and enhanced academic outcomes among affected individuals.