ENHANCED CHAMELEON SWARM OPTIMIZATION-BASED ENSEMBLE LEARNING APPROACH FOR SLEEP STAGE CLASSIFICATION FRAMEWORK

Abstract

In today’s world, every individual faces the maladies of sleep malfunction, which is diagnosed by analyzing the PolySomnoGraphy (PSG) records. The primary remedy is systematically identifying the sleep stages. But practically, the sleep stage analysis depends on the visual examination of the 30[Formula: see text]s time duration of PSG record signals. Such expertise has developed an enormous model to rectify the issues. Albeit, it exhibits effective performance, it becomes cumbersome to handle the large-scale dimension of data. Thus, the model becomes fragile when it is implemented in some sleep clinics. Even if it contains a better compatible of PSG; it subsists with limitations such as obtaining less efficiency and mitigates the robustness of the system. To alleviate this problem, a novel scheme is proposed by an Ensemble Learning Approach (ELA) for sleep stage classification using ElectroEncephaloGram (EEG) signals. Initially, the required EEG signals for the proposed model are gathered from the standard data sources. Further, the garnered EEG signals are utilized in the sleep signal extraction phase, in which the band pass filter is employed for extracting the sleep signals. The extracted sleep signal is used under the signal decomposition phase, which is accomplished by 5-level Discrete Wavelet Transform (DWT) for decomposing the sleep signals. Subsequently, the decomposed signals are given into the feature extraction phase, where the spectral features and Principal Component Analysis (PCA) are used for extracting the EEG signal features. Further, the extracted features are fed as input into ELA using Deep Temporal Convolutional Networks (DTCNs), Long Short-Term Memory (LSTM), and Deep Neural Networks (DNNs) for classifying the sleep stages. Here, the hyper-parameter tuning is engaged for enhancing the classification performance using the Modified Position Updated Chameleon Search Algorithm (MPU-CSA). The experimental analysis is conducted by distinct validating metrics for revealing the efficacy of the developed sleep stage classification model over the conventional models.