From data to pre-diagnosis: an evaluation of Machine learning models for dyslexia detection

Duque Maldonado, Alejandra

From data to pre-diagnosis: an evaluation of Machine learning models for dyslexia detection

License

Citation

Duque Maldonado, A. From data to pre-diagnosis: an evaluation of Machine learning models for dyslexia detection. 2023. handle: http://hdl.handle.net/10230/57764

This citation was generated automatically.

Abstract

Dyslexia is a learning difficulty related to the processing of language that affects approximately 10% of the Spanish population. Early detection of dyslexia is crucial to prevent academic failure or abandonment. In the present day, there are automated methods, such as machine learning methods, to process data, recognize patterns and detect pathologies. A previous study by Rello (2020) demonstrated that it is possible to apply machine learning to the detection of dyslexia. Rello created an open access tool through which, by performing a linguistic-based test, users can learn whether they might suffer from this disorder. Rello (2020) investigates the use of an algorithm that, from the processing of data, classifies users in the “Dyslexia” or “No dyslexia” classes. The goal of this thesis is to replicate the algorithm by Rello and contrast if other machine learning methods could be effective in detecting dyslexia. By training three supervised learning algorithms, we have obtained an algorithm with 92% of sensitivity for the “Dyslexia” class. More generally, this study demonstrates that other methods can also be satisfactory in the detection of dyslexia, and that the selection of parameters can influence the improvement of the results of a model.