Early Detection of Autism Spectrum Disorder in Children Using Different Machine Learning Algorithms

Abstract Autism spectrum disorder(ASD) is a neurological condition marked by impaired communication abilities, social detachment, and repetitive behaviors in individuals. Global health organization facing difficulties in establishing an effective ASD diagnostic system that facilitates precise analysis and early autism prediction. It is a scientific issue that necessitates resolution. This research presents an approach for the early prediction of children with ASD utilizing significant variables through machine learning (ML) methods. Three stages comprise the suggested technique. First, a 1250-case ASD dataset was identified and preprocessed. Five extremely effective traits with high Pearson correlation coefficient (PCC) are chosen from 10: Sex, Speech delay, Jaundice, Genetic disorders, and family history. Next, chosen ASD feature dataset through its paces using five ML techniques: Naive Bayes (NB), K-Nearest Neighbor (k-NN), Decision Tree (DT), Support Vector Machine (SVM), and AdaBoostM1 (ABM1). The proposed framework is assessed in the third phase utilizing five measurements such as accuracy, precision, predicting time, recall, and F1-score,. The findings revealed that: The NB and K-NN approaches exhibit superior accuracy rates of 99.2% and 97.2%, with minimal prediction times of approximately 0.3 seconds and 0.45 seconds, correspondingly. Conversely, the DT and AdBM1 methods demonstrate a minor decline in accuracy, achieving 94.8% and 87.6%, respectively, along with increased prediction times. Nonetheless, the SVM approach exhibits the least performance, achieving an accuracy of 80.4% with a highest prediction time of 0.84 seconds.