Investigation of artificial intelligence patterns in the diagnosis of autism Spectrum disorder (ASD)

Background and aims: Autism does not have a known cure, but early detection and interventioncan greatly improve outcomes for patients. Machine learning algorithms are being developed tohelp detect autism as early as possible, in order to ameliorate the effects of it. In this review, weexploring the machine learning algorithms and identify the best of them.Method: To conduct this review on the algorithm of machine learning to detect Autism SpectrumDisorder (ASD), an electronic literature search was performed using various online databases.The databases used for the literature search included Google Scholar, IEEE Xplore®, Science Direct,Scopus®, and PubMed. The literature search was conducted using different combinations ofsearching keywords or terms to ensure that all relevant articles were retrieved. The search termsutilized were “autism spectrum disorder,” “ASD,” “autism,” “pervasive developmental disorder,”“PDD,” “diagnosis,” “machine learning with ASD,” “mental health,” “mental illness,” “mentaldisorder,” “genetics,” “supervised learning,” “unsupervised learning,” “gene expression in ASD,”and “data mining.”The process of selecting articles involved assessing whether they met certain criteria for inclusionand exclusion. To be included, the articles had to be written in English, reviewed by peers,published mostly between ۲۰۱۵ and ۲۰۲۰, related to autism datasets, focused on classificationand feature selection of ASD, analyzed ASD and other similar disorders, related to autism withmachine learning, and discussed ASD datasets with users’ security and privacy. Articles that werenot related to ASD and similar disorders or did not meet any of the inclusion criteria were excludedfrom the selection process. Out of ۲۹۳ articles found, ۵۰ research articles were selected afterreviewing their abstracts, methods, and results. We use datasets such as: UCI ABIDE I ABIDEII NDAR AGRE NRGR GEO SSC Simons VIP. Our finding showed that ADTree, SVM, Ridgeregression ENet, LASSO, SVM, LDA, Ridge regression with ۹۸%, fNN (۹۹%), and Binary fireflyalgorithm with ۹۷% accuracy had the highest accuracy to detect the ADS.Conclusion: Efficient diagnostic performance is crucial for accurately and cost-effectively classifyingthe type of ASD, which can be improved through machine learning algorithms. However,proper feature selection and addressing critical issues such as data size and security are necessaryfor optimal classification and diagnostic accuracy.

کلیدواژه ها:

artificial intelligence ، autism ، machine learning ، computational neuroscience ، autism detection