A Machine Learning Techniques for Early Autism Spectrum Disorder Detection Through Comparative Analysis of Feature Engineering and Classification Models

Autism spectrum disorder affects how someone perceives and interacts with others, which can lead to issues with social interaction and communication. It is a condition related to brain development. It is preferable to declare that this disorder cannot be cured than to say that it can be averted at an early stage. This article provides a useful framework for evaluating several ML techniques for early ASD diagnosis. The Quantile Transformer (QT), Power Feature Scaling (PFS), Power Normalizer, Max Abs Scaler, and Transformer (PT) are employed in this proposed model. Eight effective classifiers including AdaBoost (AB), Gaussian Nave Bayes (GNB), XGBoost, Support Vector Machine (SVM) Ensemble method, K Nearest Neighbor(KNN) , Random Forest (RF) and Decision Tree(DT) are then used to categorise the feature-engineered dataset. Comparing categorization outcomes produced using different evaluation measures like Recall, Precision, F1 score, Kappa score, Mathews Correlation Coefficient (MCC) and Accuracy. After comparing the results of several ML classifier models, it was found that the ASD may be easily recognised using the ensemble technique since it produces the highest accuracy—99.22%—on feature-scaled ASD datasets.