Machine Learning-Based Web Application for Early Detection of Autism Spectrum Disorder

This study outlines the development of a web application leveraging machine learning for the early diagnosis of autism spectrum disorder (ASD). The proposed system focuses on creating a predictive model using data from the Q-CHAT-10 questionnaire to identify patterns associated with ASD. Multiple classification algorithms were evaluated, including Logistic Regression, Support Vector Machines (SVM), Decision Trees, Random Forest, XGBoost, K-Nearest Neighbors, and Gaussian Naive Bayes. To enhance system performance, techniques such as hyperparameter tuning with Grid Search and class imbalance management using SMOTE were employed, resulting in improved model accuracy and generalization. The outcomes surpassed the established benchmark of 85% across key metrics, including precision, accuracy, recall, F1 score, and AUC-ROC. Notably, the optimized SVM model demonstrated superior performance, achieving an accuracy of 98.80%, precision of 93.27%, recall of 90.11%, F1 score of 94.25%, and an AUC-ROC of 97.93%. This technological solution shows significant potential for integration into clinical settings, facilitating early diagnoses and providing effective decision-making support for healthcare professionals. Furthermore, the results emphasize the application's effectiveness in resource-limited environments, representing a meaningful advancement in ASD care within clinical contexts.