Machine Learning Based Early Detection of Autism and Severity Classification

Abstract

Autism Spectrum Disorder (ASD) is a neurodevelopmental condition that appears in early childhood and affects social interaction, communication, and behavior. Early identification of ASD in toddlers is crucial. Intervention during the early developmental window, which spans from 12 to 36 months, has been shown to significantly improve learning, communication, and adaptive skills. The current ASD screening system mainly relies on manual questionnaires and clinical observation; these methods are time-consuming. The machine learning based early autism detection model focuses on creating an automated early screening system for ASD. The proposed system aims to identify children at high risk of ASD by analyzing behavioral screening responses and related demographic information in a reliable and objective way. This system is developed using Python programming with relevant machine learning algorithms to support early and trustworthy risk identification. Multiple supervised machine learning models, including Multilayer Perceptron (MLP), Naive Bayes, Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), and AdaBoost, are trained and evaluated. These algorithms work together to detect autism and indicate severity levels such as mild, moderate, and high. This approach reduces human error and delays in detection. It also supports early clinical referral and timely intervention, leading to better developmental outcomes and quality of life for children and their families. The proposed system removes the need for repeated clinical visits and costly diagnostic procedures, thereby lowering the overall costs of screening and assessment.