Neutrosophic Weighted Support Vector Machine for Autism Spectrum Disorder Detection in Children

Abstract

Introduction: Many studies have been introduced in data analysis and machine learning algorithms, especially in the medical and healthcare field, such as Autism Spectrum Disorder (ASD). Aim of Study: The study aimed to find an effective way to classify children with autism spectrum disorder based on the concept of Neutrosophic sets that fits the nature of real data, which may be incomplete or with noise. The proposed method calculates a set of values of weights based on the concept of Neutrosophic membership using Neutrosophic clustering by kernel method, these weights are passed to the Neutrosophic Support Vector Machine to find the optimal hyperplane. Research Methods: The proposed method was evaluated using the UCI Machine Learning Repository data set containing 292 cases and 21 attributes, and it has noise and is incomplete by up to 15%. We compared the proposed method with the Fuzzy Neutrosophic Support Vector Machine (FNSVM) and Support Vector Machine (SVM) methods, conducted using a python programming language. Experimental results: confirm the outperform of the proposed method with an accuracy of 99.98%. In addition, the result of classification accuracy of this method applied to raw data was compared with other methods such as CNN, ANN, and Stochastic GSS by using the same data but after application of initial processing of missing values and selection of features. Discussion: The comparison results prove the superiority of our method, which is based on Neutrosophic theory.