<![CDATA[Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder that affects communication, social interaction, and behavior in children. Early detection of ASD is crucial because early intervention has been proven to improve children's adaptive abilities. However, the screening process in Indonesia is still conducted conventionally through observation and interviews, which require considerable time and are highly subjective. This study aims to implement a prototype web-based ASD screening system using a Support Vector Machine (SVM) optimized through feature selection and hyperparameter tuning. The dataset used is the Autism Child Screening Data (n=292) with 22 initial variables covering demographic data and Q-CHAT-10 screening results. The class distribution is relatively balanced, with 151 Normal (51.7%) and 141 Autism (48.3%) samples. The research stages include data preprocessing, feature selection using four methods (correlation, Mutual Information, ANOVA F-score, and Permutation Importance), hyperparameter tuning with GridSearchCV 10-fold cross-validation, model training, evaluation, and web-based system implementation using Streamlit. The results show that the SVM model with the RBF kernel, parameters C=10.0, gamma=0.01, and class weight {0:1.0, 1:3.0}, achieved an accuracy of 94.83%, a precision of 100% for the Normal class and 90.32% for the Autism class, a recall of 90% for Normal and 100% for Autism, an F1-score of 0.95 for both classes, and an ROC-AUC of 0.9940. The model is implemented in a web prototype that can be used as an early screening aid for healthcare professionals. This research contributes to the development of a machine learning-based decision support system integrated with clinical knowledge through the Q-CHAT-10 cut-off post-processing rule.]]>
Copyrights © 2026
