A novel hybrid deep learning model using MEResNext for autism spectrum disorder detection.

A neurological disease, named autism spectrum disorder (ASD), is portrayed through divergence with social interaction, communication, and repetitive activities. As heredity is an important source, initial identification and treatment can diminish the need for wide examinations and expensive medical procedures for those with ASD. Prolonged effects and severity can be avoided at a timely determination of ASD. This study develops a hybrid deep-learning method for ASD detection. Pre-processing, feature selection and ASD detection are three steps involved. In the pre-processing phase, input data is passed in which noise and artefacts present in the data are removed using Yeo-Jhonson transformation. Next, by integrating Double exponential Smoothing (DES) in Elk Herd Optimizer (EHO), Double exponential Smoothing-Elk Herd Optimizer (DeSEHO), is proposed to select features. At last, the features that are selected are subjected to the ASD detection phase. In this stage, the Moments Embedding ResNeXt (MEResNeXt), which is the combination of Moments Embedding Network (MoNet) and ResNeXt, is used for ASD detection. The results of the research show that MEResNeXt performed better in terms of traditional models which show accuracy of 95.3 %, sensitivity of 96.5 %, and specificity of 94.8 %.