Autism Screening Prediction Based on Multi-kernel Support Vector Machine

Autism is a group neurodevelopmental disorder that affects a person's ability to communicate, engage with others, and act in specific ways. Screening and early prediction of autism are important in supporting the proper diagnosis and effective early intervention. However, early detection of autism spectrum disorder (ASD) can be challenging since it is behavior-based and more complicated to identify biologically. Our classification system uses ten questions from the Autism-Spectrum Quotient (AQ) Test and four specific subject data features. Support Vector Machine (SVM) is a machine learning method that has shown effectiveness in prediction and classification. However, when the investigation data has complex non-linear characteristics, SVM with a single kernel may not perform as well. Our study proposed a multi-kernel SVM method to improve the model's accuracy. The classification evaluation parameters, including accuracy, precision, recall, and F1 score, were applied to evaluate the prediction model's effectiveness. Our experiment results indicate that the Multi-kernel SVM approach has a robust performance in predicting autism. Furthermore, our proposed method's performance is outstanding, with an accuracy of 0.976; the normal class can be distinguished with high precision, recall, and f1-score, whereas the ASD class has an ideal precision of 1.00.