Identification of Autism Spectrum Disorder Using Modified Convolutional Neural Network (MCNN) and Feature Selection Techniques

Abstract

ASDs (Autism Spectrum Disorders) are neurological disorders that impact people’s lifetime capacity to connect or interact with others. Autisms are behavioural diseases where symptoms often erupt in the first two years. The problems may bein at very early ages in patients of ASDs and extend into adolescence. The previous research developed a method called AGWO (Adaptive Grey Wolf Optimisation) to find the most important characteristics and effective classification methods in ASD datasets. However, because of the high training complexity of SVM (Support Vector Machines), this model employed a single classification-based prediction method, which is not appropriate for classifying big data sets. In order to effectively identify ASD, this study effort developed M-AECNN (Modified Auto Encoder Convolution Neural Networks) based classification and dimensionality reduction algorithms. In this work introduced a AGWO to identify the most significant attributes and efficient classification techniques in ASD datasets. Initially the SMOTE based pre-processing approach is applied for removing the irrelevant data in ASD dataset. Subsequently, The AGWO algorithm keeps going through this process until it finds the characteristic that has the lowest classification recall and accuracy. Finally, M-AECNN-based classification is used to determine if a dataset instance is ASD. The highest performing classifier for these binary datasets was identified by experimental examination of ASD datasets from toddlers, children, adolescents, and adults, taking into account recall, precision, F-measures, and classification errors. The dragonfly optimisation is introduced in this paper for optimising the overfitting in AECNN to increase performance.