Polynomial Regressive Quadratic Gradient Optimized Deep Belief Classifier for Autism Spectrum Disorder Identification

Abstract

ASD is neurological chaos which affects individual's brain development, influencing their communication skills, and behavior. Premature recognition as well as interference considerably enhances results for individuals through ASD. Interventions may include behavioral therapy, and so on. ML methods have applied to classification as well as diagnosis of ASD to analyze large datasets.  However, conventional models have faced major challenges, leading to inaccuracies and increased time consumption. To address this challenge, a novel deep learning model called the Polynomial Regressive Quadratic Gradient Optimized Deep Belief Classifier (PRQGODBC) is developed to enhance accuracy of ASD identification. The proposed deep belief classifier is kind of DL ANN with numerous layers of nodes (neurons) organized into an input layer, one or more hidden layers, output layer. These processes are integrated into the proposed Deep Belief Classifier to enhance accuracy and minimize time in Autism Spectrum Disorder (ASD) identification. During data acquisition stage, patient data points are gathered as of database. These collected data points are then fed into input layer of deep belief classifier. Input is subsequently transferred to the hidden layers of deep belief classifier, where preprocessing, feature selection, categorization are performed. In preprocessing steps, two processes are considered such as handling missing data and removing noisy data using polynomial regression-based data imputation technique and greatest normalized residual test. After preprocessing, feature selection is performed by Quadratic Discriminant Analysis to choose the most relevant attributes from dataset. Through chosen features, classification tasks are carried out using Tucker’s congruence coefficient and provide classification outcomes for normal or autism identification. After classification, a fine-tuning is applied to reduce error rate by optimizing hyperparameters using the stochastic adaptive gradient method. Finally, accurate classifications of ASD patients are attained at output layer. An experimental assessment is conducted through various factors. Obtained outcomes show that PRQGODBC is more efficient in achieving higher accuracy while maintaining time compared to existing approaches.