A Hybrid Approach for Autism Spectrum Disorder Classification

Autism spectrum disorder (ASD) is a neurological condition that can be devastating to the social functioning of the affected person. It is attributed to a range of symptoms that include troubles in social interaction, difficulty in expressing themselves and repetitive pattern filled behavior. People with autism have a unique behavioral pattern and the severity of the disease may vary across individuals, the causes for which are not known. The prevalence of ASD is increasing globally and early diagnosis of the disorder can lead to substantial behavioral improvements. Machine learning techniques are widely used in the health care domain for medical diagnosis. The study focuses on applying machine learning ensemble techniques to autism adult data sets to predict autism in adults. The UCI Machine Learning Repository's Autistic Spectrum Disorder Screening Data for Adult was used for the experiment purpose. The hybrid approach makes use of rough set algorithms for feature selection using Rosetta rough set tool and Adaboost with decision stump for classification using Weka data mining tool. Classification accuracy was high when the dataset was selected based on the reducts generated by Genetic algorithm. Results indicate that the proposed hybrid model improves the performance of autism data classification.