Developing a Classification Framework for Autism Spectrum Disorder Detection in Children Using Machine Learning Approach

Human behaviors are crucial in shaping identity, influencing social interactions, and determining how individuals perceive and adapt to their environment. Autism spectrum disorder is a neurological developmental condition marked by difficulties in behavior, communication, and social interaction. Early diagnosis and access to therapy services are essential for improving outcomes. However, current screening methods encounter challenges, including delays, lengthy processes, dependence on highly trained healthcare professionals, and significant costs. This study aims to enhance the efficiency and accuracy of ASD diagnosis in children aged 4-11 years by employing optimized Feature Selection techniques. Data samples were sourced from the UCI Machine Learning repository. We developed and evaluated Machine learning models, including K-Nearest Neighbors (KNN), Naïve Bayes (NB), and Artificial Neural Networks (ANN) with different optimizers to facilitate timely ASD screening in Children. The results indicate that the KNN model, using a subset of features, achieved the highest precision and sensitivity (0.96) with an 80:20 split, outperforming other models. These findings highlight the potential of machine learning in simplifying ASD screening, making it more accessible, efficient, and cost-effective. By reducing complexity, this approach may help overcome current clinical hurdles and improve early access to treatment for children with ASD.