An Improved Feature Selection Algorithm for Autism Detection

Abstract

Autism Spectrum Disorder (ASD) is one of the most common acute neurodevelopmental disorders. It is associated with the development of the brain. ASD severely affects a child's physical and mental health. ASD detection at an early age is challenging as its symptoms come after two years. Each ASD patient has a different set of symptoms (features). In recent years, machine learning has offered a new potential solution for the detection of Autism. The effectiveness of the machine learning models depends on the dataset's features. This paper proposes a feature selection algorithm (which is based on feature correlation and ranking) for early ASD detection on the clinical ASD dataset. The performance of the feature selection algorithm is compared with different machine learning algorithms (LR, GBC, AdaBoost, and DT). The result shows that 5 out of the 30 features with a Logistic Regression model are sufficient to detect Autism with 98.18% accuracy, 98.16% sensitivity, and 98.16% precision. The result also shows that the Gradient Boost achieves 98.18% accuracy with 5 features, and the AdaBoost achieves 97.10% accuracy with 5 features.