Authors - Usman Ali, Shaiza Shahid, Misbah, Shahzad Ali, Sheraz Aslam, Kainat Mustafa Abstract - Autism Spectrum Disorder (ASD) is a neurocognitive condition that affects social connection, expressive abilities, and behavioral patterns. This study aims to diagnose ASD using machine learning (ML) techniques based on behavioral and demographic data. The dataset sourced from Kaggle includes features such as age, gender, ethnicity, social interaction behaviors, and responses to standardized screening questionnaires. To improve accuracy, various ML models are implemented, including NB, AdaBoost, SVM, LightGBM, KNN, and XGBoost. Additionally, advanced Ensemble Methods are applied, such as Hybrid Kernel SVM, a boosting classifier, and a stacking classifier. Among all methods, the stacking classifier delivered the best performance, achieving 99.74% accuracy with the minimum MSE of 0.2. This result supports the development of an intelligent system to aid in early autism diagnosis and intervention planning. Future research could enhance this approach by integrating neuroimaging biomarkers, generative AI, or more advanced ML frameworks, aiming for even greater precision and robustness in a real-world clinical setting.
