Towards the Development of Explainable Machine Learning Models to Recognize the Faces of Autistic Children

Machine learning with image classification has shown promise in supporting the detection ofautism in children, but the development of explainable models is still lacking. Thus, our studycompared the use of two algorithms to explain why facial images are categorized as autistic ornot. First, we trained and tested different models on the Autistic Children Facial Image Data Setto identify the one that produced the highest accuracy. Following the identification of the bestmodel, the analyses compared two methods to examine explainability: Local InterpretableModel-agnostic Explanations (LIME) and Randomized Input Sampling for Explanation of black-box models (RISE). Overall, the best model, ViT_Huge_14, produced an accuracy of 92% andLIME resulted in more explainable models than RISE. Albeit promising, researchers mustconduct further studies to examine the generalizability of the results prior to recommendingfacial image classification as a component of a multimethod approach to screening anddiagnosis.