Classification models for improved prediction of autism spectrum disorder: behavioural and facial feature analysis using machine learning approach

Autism spectrum disorder (ASD) is a prevalent neurodevelopmental condition that significantly affects social relationships, interaction, behavior, learning capabilities, and knowledge acquisition. Despite its global prevalence and progressive nature, timely diagnosis remains challenging. This work aims to develop a machine learning classifier model to accurately predict the likelihood of ASD by utilizing dataset comprising both behavioral and facial image features from children. ASD symptoms typically appear and develop more between ages 3 and 11, with early signs often misinterpreted as other conditions, such as intellectual disability, which delays accurate diagnosis. Diagnosing ASD involves evaluating developmental history and observing behaviors, and processes that require considerable expertise and time. Recently, machine learning and deep learning have shown promise in ASD research, particularly in enhancing diagnostic accuracy through data analysis. The models employed in this study include Logistic Regression, K-Nearest Neighbours, Support Vector Machine, Extra Tree, Random Forest and Artificial Neural Network. Using the image-based data, the SVM model achieves 70% accuracy in image classification, while the ET and LR models yield 100% accuracy using the AQ-10 screening dataset, outperforming other models analyzed. This research demonstrates the potential of machine learning models in improving the accuracy of ASD diagnosis by integrating behavioral and facial image data. The findings suggest that employing diverse dataset and advanced algorithms can significantly enhance the diagnostic process.