Machine learning data-based approaches for autism spectrum disorder classification utilising facial images

Autism Spectrum disorder (ASD) is a neurodevelopmental disorder with a broad range of symptoms which, however, differ in severity. Regarding early autism detection, parents or experts may notice early indications of the disorder before a child turns one year old although the symptoms usually become more noticeable by the time the child is two or three years old. Thus, it is of high importance to employ a cutting edge method for ASD detection in years to come. In this study, a Kaggle dataset including 3014 images of ASD and non-ASD children was used for ASD recognition from facial images of children. A crucial part of the research is the pre-processing stage and concoction of the dataset. Ensemble learning techniques, such as Random Forest Classifier, Gradient Boosting Classifier, and Support Vector Machines were implemented. Regarding the outcomes derived from the referenced algorithms, gradient boosting proved to be the most sufficient algorithm with an accuracy of approximately 90% accuracy in the best occasion coming in two times faster than SVM also reaching the same result and outperforming the Random Forest Classifier. By integrating powerful algorithms with precise pre-processing procedures, the present study intends to contribute to autism detection, early intervention, and consequently to a better quality of life for ASD individuals.