Autism spectrum disorder detection and classification using chaotic optimization based Bi-GRU network: An weighted average ensemble model

Abstract

Autism Spectrum disorder (ASD) is a neurological disorder in which a person suffers from lifetime issues like social communication and interaction with other individuals. This disorder starts in childhood and leads to adulthood. So, a person’s entire life is fully affected by this disorder. An early diagnosis is crucial to reduce the ASD’s symptoms and also to enhance the ASD patient’s life. The manual screening takes more time and tedious process. Hence, Electroencephalogram (EEG) signals are utilized for the brain’s electrical activity recording process. The EEG signals are time-varying and non-stationary signals and also various methods are utilized to extract features using the EEG signals for the classification process. In this paper, we propose a novel hybrid ensemble model which is the ensemble of ResNet101 and a Bidirectional Gated recurrent unit (Bi-GRU) with a Weighted Average Ensemble (WAE). The preprocessing is initially carried out to remove undesirable aspects from the input signal, such as noise, computational burden, etc. A convolutional neural network-based ResNet model is utilized for the classification and detection of input data. Bi-GRU neural network learns data from two different sources such as forward and backward to provide extremely precise predictions. We integrate ResNet and Bidirectional Gated recurrent unit (Bi-GRU) to produce accurate classification results and this technique is optimized by using the Chaotic Henry Gas Solubility Optimization (CHGSO) algorithm. To enhance the efficiency of the proposed method we compare the four underlying techniques such as the Support vector machine (SVM), K-Nearest Neighbor (KNN), Modified Grasshopper Optimization Algorithm- Random forest (MGOA-RF), and Deep Neural Network (DNN), and the hybrid model is evaluated with dataset EEG microstates dataset. Accuracy, precision, sensitivity, specificity, F1-score, and MCC are utilized for assessing the classification performance more accurately. Our Hybrid ensemble model attains Sensitivity of 98%, 99% higher Specificity, 98% F1-Score, MCC of 99%, Accuracy of 98%, and Precision of 99%.