Neuro Insight AI: Intelligent Early Detection of Autism Spectrum Disorder

Abstract

This research presents NeuroScan Artificial Intelligence, a comprehensive machine learning framework designed to enhance early autism spectrum disorder (ASD) detection through advanced predictive analytics. Traditional ASD screening methods relying on manual questionnaire scoring often lack accuracy and adaptability across diverse populations. Our solution employs an ensemble of six machine learning models (XGBoost, Random Forest, Logistic Regression, SVM, Gradient Boosting, and Neural Networks) trained on clinically-relevant engineered features, including domain-specific behavioral scores, age-adjusted metrics, and biological risk factors. The system processes input from standard A1-A10 screening questionnaires, transforming them into sophisticated predictive features through automated preprocessing and feature engineering pipelines. Our results demonstrate exceptional performance with Logistic Regression, achieving 98.2% accuracy and 0.98 AUC score, significantly outperforming traditional screening methods. The framework incorporates stratified cross-validation, robust handling of class imbalance, and comprehensive evaluation metrics to ensure reliable predictions. Beyond binary classification, NeuroScan AI provides probability-based risk stratification, domain-specific behavioral analysis, and evidence-based clinical recommendations through an intuitive interface featuring real-time visualizations, including probability gauges, feature importance charts, and interactive analytics. This approach bridges the gap between computational efficiency and clinical utility, offering healthcare professionals an objective, scalable tool for early ASD identification while maintaining interpretability through feature importance analysis and transparent probability scoring. The system's modular architecture allows continuous learning from new data, making it adaptable to evolving diagnostic criteria and diverse demographic populations.