An Innovative Gated Graph Recurrent Neural Network in Toddler Autism Prediction

In medical diseases like autism spectrum disorder (ASD), key features from symptoms, such as autistic symptoms, should be understood for early diagnosis at the time of clinical screening process. Since this will directly impact model precision and efficiency, choosing appropriate features is one of the most important steps in model developments for such medical applications as ASD screening. This framework aims to examine the integrated relationship of autism data with a view to accurately predicting the disorder in toddlers. Preprocessing and autism forecasting are two phases designed within this framework. First, the Var-SS normalization technique is proposed in the preprocessing stage itself for removing unsolicited missing values from the database. Then, the Gated Graph Recurrent Neural Network (GGRNN) technique is presented to envisage the existence and non-existence of autism disorder in toddlers. The developed framework is processed under the Python platform and the freely accessible toddler autism detection database is utilized for the experimentation process. Various computational metrics like F-measure, accuracy, recall, MCC and kappa coefficient (KC) are evaluated and distinguished from different studies. The overall accuracy of 98.52%, recall of 98.08%, F-measure of 98.86%, KC of $\mathbf{97.06\%}$, and MCC of 97.65% are obtained by the developed technique against various conventional studies on forecasting the autism disorder in toddlers.