Multi-Scaler Fusion Framework for Pediatric Autism Prediction Using Ensemble Machine Learning and Feature Attribution

Abstract

In this study, we provide a machine learning framework for ASD early diagnosis that makes use of four remarkable feature
scaling methods and eight main machine learning algorithms. The suggested design makes use of four outstanding feature
scaling (FS) methods: quantile transformer (QT), power transformer (PT), normaliser, and max abs scaler (MAS). For the
feature-scaled datasets, eight trustworthy machine learning methods are utilised: AdaBoost (AB), Random forest (RF),
decision Tree (DT), k-Nearest neighbours (KNN), Gaussian Naïve Bayes (GNB), Logistic Regression (LR), guide Vector
system (SVM), and Linear Discriminant analysis (LDA). The study utilises four standard datasets on ASD: infants, young
adults, children, and adults. Basic function selection approaches and primary type algorithms for every ASD dataset are
observed. To get the most reliable results, use the normaliser FS on younger children and the QT FS method on older adults
and teens. In order to rank the important attributes in order of importance, four feature selection techniques (FSTs) are
used: information gain attribute evaluator (IGAE), gain ratio attribute evaluator (GRAE), relief F attribute evaluator (RFAE),
and correlation attribute evaluator (CAE). These FSTs are used to assess the risk factors for autism spectrum disorder (ASD).
Results from using the proposed paradigm for early ASD diagnosis are better than those from using current methods.