Developing an AI-driven Healthcare System for Predicting Autism Spectrum Disorder

Behavioral observations have traditionally served as the foundation for diagnosing autism spectrum disorder (ASD). However, these conventional diagnostic methods often present challenges, including potential inaccuracies and significant time demands. The integration of technological screening tools and machine learning algorithms with standard behavioral observations has the potential to enhance the assessment and diagnostic processes for ASD, leading to more efficient and accurate outcomes. Accurate and reliable classification of ASD is essential in medical diagnosis. In this study, the performance of the random forest (RF) and K-nearest neighbors methods was evaluated for ASD detection using a public dataset. This dataset, collected from the Kaggle repository, includes 704 samples of adult autism screening, with 20 attributes designated for future research, especially in identifying key autistic symptoms and refining ASD categorization. The dataset features 10 behavioral traits (AQ-10-Adult) and 10 personal characteristics that have proven effective in distinguishing ASD patients from controls in behavioral science. Data preprocessing involved encoding, feature selection, and data partitioning. The RF model achieved a high accuracy of 99.29%. Our findings illustrate the efficacy of the proposed RF model in accurately diagnosing ASD through comprehensive data analysis and performance metrics. This approach facilitates a more rapid identification of the condition, thereby enhancing the overall detection of ASD.