Pre-Diagnosis of Autism in Children Using Distributed Learning

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition with varied symptoms affecting social communication, behavior, and sensory processing. Diagnosing ASD in children requires a thorough assessment of their behaviors, making it a nuanced process. This study explores the use of machine learning (ML) algorithms—logistic regression, decision tree, Naïve Bayes, support vector machine, and random forests—to streamline ASD diagnosis in children. Utilizing a modified Privacy-Preserving Aggregation of Teacher Ensembles (PATE) Framework, it combines small datasets from multiple practitioners and various ML approaches while preserving data privacy through encryption. Our results show that the PATE approach enhances the sensitivity, specificity, and accuracy of ASD diagnosis in children, compared to individual methods, while ensuring data privacy. This study advances collaborative predictive modeling in psychology, helping practitioners with small datasets work together effectively.