A Hybrid Approach for Prediction of Autism Spectrum Disorder using Machine Learning Algorithms

Autism Spectrum Disorder, commonly known as ASD, is a type of neurodevelopmental disorder that is typically identified during early childhood. Early parental observation is crucial in timely diagnosis, facilitating prompt interventions and therapies the diverse range of symptoms and varying severity levels of ASD present significant diagnostic challenges. While many previous studies have focused on diagnosing ASD in adults, early detection remains paramount despite the disorder's incurable nature. In this study, a predictive ASD model for toddlers' datasets was presented, utilizing ensemble learning methods such as voting. This approach complements conventional learning methods. The predictive analysis of ASD is carried out in two stages, first, ASD diagnosis is conducted utilizing a range of classification algorithms, such as Random Forest (RF), Bernoulli Naive Bayes (NB), K-Nearest Neighbors (KNN), Decision Tree (DT), and Support Vector Machine (SVM) with Radial Basis Function (RBF) kernel on the dataset of toddlers. To improve the performance of the ASD prediction model in the final stage, the voting ensemble learning approach was used in conjunction with three classifiers from the base above learners. The findings show that the proposed ensemble-based prediction models outperform basic Machine Learning (ML) approaches regarding accuracy, recall, F1 score, and precision, making them more useful for early-stage ASD diagnosis in toddler datasets.