Empowered Deep Belief Network Learning fused with Shuffled Frog Leaping Algorithm for Autism Spectrum Disorder Prediction among Children at its Early Stage

Abstract

Despite the ongoing rise in the frequency of autism spectrum disorder (ASD), prompt intervention and improved results depend on efficient early screening.For prompt treatments to enhance outcomes and help children reach their full potential, preliminary identification of Autism is essential.The main problem statement of this paper is to expedite Autism diagnoses by providing a machine learning system that uses different machine learning algorithms that lead to the make Autism predictive model with most possible accuracy. There are many machine learning models are developed by researchers to detect the autism at its early stage, but the problem of class imbalance greatly affects the performance and challenging issue. After the emergence of deep learning methods, the voluminous dataset is handled very effectively by this large network with depth layers. Still, the problem of overfitting due to class imbalance affects the accuracy rate in autism detection. Hence, this paper highlights the issue of overfitting while using deep learning algorithm which occurs mainly due to inappropriate assigned of hyperparameter values. The proposed model overcomes this issue, by adopting Shuffled Frog Leaping Algorithm (SFLA) to assign the optimized values to the hyperparameters of Deep Belief Network based Learning Model (DL). The fitness value evaluation of SFLA is used for learning rate and weight parameter assigned in DBN, while the conventional DBN follows it in a random approach. The UCI ML repository and Kaggle provide the data used to apply the approaches.The simulation results proved that the proposed EDL-SFLA produced the highest accuracy rate of 98.2% in prediction of autism among children compared with other conventional models.