Early Autism Detection: A Machine Learning Approach Using Toddlers’ Data

The paper explores machine learning algorithms for Autism Spectrum Disorder (ASD) detection, including KNN, RF, Decision Trees (DT), LDA, SVM, Logistic Regression (LR), hybrid KNN+RF, KNN+DT, KNN+LDA, KNN+SVM and KNN+LR.In this study, we developed toddler prediction models for Autism Spectrum Disorder (ASD) by analyzing four distinct datasets using eleven machine learning (ML) classifiers and various feature selection (FS) methods. Among the individual algorithms, KNN emerged as the top performer, achieving the highest accuracy (98.62%) and MCC (0.9871). Motivated by KNN’s strong performance, we hybridized it with other models to enhance predictions. The KNN+RF combination proved to be the best, achieving 99.29% accuracy, along with high precision, recall, F1-Score, and the lowest LogLoss (0.8673), indicating robust and balanced predictions. Other combinations like KNN+LR also performed well, while KNN+DT and KNN+LDA showed reduced effectiveness due to higher LogLoss and lower MCC. These findings highlight the potential of hybrid models, with KNN+RF offering the most reliable solution for ASD prediction.