Article Preview
Top1. Introduction
In this work a machine learning-based knowledge generation system has been proposed, that can predict the presence of the Autism Spectrum Disorder(ASD), based upon the features learned by the machine learning classifier, built by using the data of the traits of the previously diagnosed individuals who were affected with ASD.
Autism spectrum disorder (ASD) is a medical state caused by the abnormal development of the features of the brain; it impacts the behavior and communication of an individual. The individuals diagnosed with ASD, sometimes also follow characteristic repetitive actions. The term spectrum has been used to describe this condition, as it covers a range of medical conditions, which were earlier known by different names like Rett's Disorder, Autism, Childhood Disintegrative Disorder, Asperger Syndrome, etc, but all these disorders had some overlapping features, therefore they have been clubbed together and the term ASD has been coined(Munda, 2015).
ASD traits are noticeable in individuals by the age of 24 months i.e. when one starts to develop socially. Although the symptoms are initially ignored, thinking that each individual has distinct timelines for the social development. But soon it is realized that some medical intervention is required, when the society start pointing at some specific behavioral patterns. It is at this point of time, the parents or the caregivers, report to the pediatric and the medical intervention is started, to analyze the reason behind the abnormal traits. Although no cure or remedy has been proposed so far, to cure this alignment, but still early medical intervention can improve the life of the subject and the others around them. (Corsello, 2005).
The major symptoms observed in the ASD affected subjects is, that they are very less interested in others, have limited vocabulary, often show less reaction to pain, may have near-average intelligence, excellent gross motor skills and weaker fine motor skills, etc. Each individual may have varying degree of these symptoms. In some cases, the individual does not show any of the above weakness, but over some time, these skills may fade away or they may even lose these abilities. Therefore, the solid traits of ASD can only be ascertained after the age of 2 years.
The reason for ASD is not yet known, but it is apprehended that neurological development is reduced, due to the mutation of the genes, which may be due to inheritance or spontaneous mutation. But some researchers feel that there may be some environmental factors like infections due to virus, pollution, etc or due to some complications during the pregnancy that might result in the onset of ASD.
ASD evaluation is a tough task, as each individual is affected differently and to a different extent. Moreover, there is not any single test that can measure the presence or absence of ASD. The diagnosis of ASD is carried out in two phases, in the initial stage, the behavioral and growth is observed and measured. If the positive indicators are seen in these two fronts for ASD, then the second phase is initiated in which the through checking is done employing genetic testing, vision, speech and hearing test, along with neurological response monitoring.
The diagnosis of ASD, is dependent upon many fuzzy factors, in the beginning it was generally carried by the medical professionals alone. But with the advancement in Artificial Intelligence and machine learning-based techniques, these days, researchers are working to develop automated machine learning-based approaches to diagnose the subjects for the presence of ASD(Bone et al., 2016).
In this work an effort is being made to develop a machine learning-based technique that helps to diagnose the individuals affected by ASD or not. The paper has been organized as: the first Section presents the introduction, Section 2 discuss the materials and methods, Section 3 presents the results of the proposed method and compares them with various state of art machine learning-based ASD diagnostic methods, Section 4 discusses the proposed method and finally Section 4 concludes the paper.