Autoencoder-based improved deep learning approach for schizophrenic EEG signal classification

In this paper, deep-stacked error minimized extreme learning machine autoencoder (DSEMELMAE) and sine–cosine monarch butterfly optimization-based minimum variance multikernel random vector functional link network are integrated to recognize the schizophrenia electroencephalogram (EEG) data. The unconventional DSEMELMAE network is modelled to derive very unique unsupervised attributes out of the brain signals and employ as inputs to the proposed supervised SCAMBO-MVMKRVFLN classification methodology to recognize accurately by minimizing the mean-square error for identifying schizophrenia data with encouraging accuracy. The DSEMELMAE-SCAMBO-MVMKRVFLN integrated approach is assessed over benchmark EEG databases. The proposed approach is compared with many related RVFLN-based deep learning approaches and many state-of-the-art methods and found to be the outperformer among all the methods, and this approach is highly accepted owing to faster learning speed, better computational simplicity, good generalization capability, outstanding classification accuracy, and small event identification time. The classifier MVMKRVFLN is unique as it classifies the signal with advantages such as the regularization of the randomization, computational economy, less training expenses, the direct inverse along with minimum reconstruction error. The KRVFLN uses multiple kernels such as wavelet, tan hyperbolic and multiquadric to improve the classification performance. The effectiveness of the proposed method is verified by examining three publicly available schizophrenic EEG datasets such as Poland, Kaggle and Moscow datasets and achieved classification accuracies with 99.989%, 95.012% and 96.69%, respectively. The recognition capability, simplicity and robustness of the proposed methodology prove the outstanding overall performances of schizophrenia recognition and diagnosis in comparison with other state-of-the-art approaches and different learning approaches.