Machine Learning Approach for Diagnosis of Autism Spectrum Disorders

Abstract

Currently affecting around 1 in 68 children, autism spectrum disorder (ASD) is a psychological disorder that is mostly diagnosed by the age of five or later. However, the catch lies in the fact that increased delay in the diagnosis leads to increase in complexities and cost of the treatment. ASD inhibits the patient from interacting with the society. Reduced social interaction and aberrant behavior are the cardinal symptoms of ASD and are usually detected in children by the age of two, when they are still in their developing phase. Thus, ASD is included in the list of developmental disorders which comprises attachment disorder, attention deficit/hyperactivity disorder, etc. Referred to as autism spectrum disorder, this umbrella term includes various types such as autistic disorder, Asperger’s syndrome, and pervasive developmental disorder. These three categories are based on the severity of autism spectrum. Autistic disorder, commonly referred to as autism, itself affects around 2 million people in India. Apart from this, while staggering statistic shows that ASD affects at least 70 million of the individuals worldwide, its diagnosis still remains an abstruse task. To ameliorate this detection process, the paper aims at finding the best machine learning algorithm for classification of the dataset into whether the person is suffering from autism or not. Besides, if the person does not show symptoms of autism, then the target is to detect whether the patient is vulnerable to any other types of ASD as discussed before. The intention of the proposed methodology is not to replace the presence of medical personnel in the process of diagnosis, but only to provide assistance and corroborate with the opinion of the concerned doctor.