Enhancing Autism Disease Classification Using a Hybrid GA-KNN Approach for Feature Selection

These days, autism spectrum disorder (ASD) is considered to be one of the most common health problems. A person with ASD is said to have a neurodegenerative disorder that affects their ability to engage and communicate. Using screening tests to identify autism symptoms is highly costly and time-consuming. The development of machine learning (ML) and artificial intelligence (AI) has made it possible to predict autism symptoms early on. Early detection of autism symptoms is now feasible due to advancements in AI and ML. The main contribution of this study is to address the high dimensionality problem by proposing a new hybrid feature selection approach using the genetic algorithm (GA) with K-nearest neighbor (KNN). This hybrid approach employs the GA-KNN to identify the ideal feature subset that enhances the performance of autism classification. The model uses three classification algorithms: Support Vector Machine (SVM), Random Forest (RF), and Convolution Neural Network (CNN) to compare their impact on classification performance and accuracy based on the proposed approach based on three ASD datasets for adults, adolescents, and children. The experimental results show that: 1) the hybrid GA-KNN approach used in the feature selection step can reduce the total number of features from 21 to 9 and increase the classifiers accuracy by 9% to 17%; 2) out of all the tested classifiers, the CNN classifier performed the best in classification, yielding the highest accuracy of 97.5%.