An Advanced Circular Adaptive Search Butterfly Optimization Algorithm for the CNN-based Sleep Apnea Detection Approach

Abstract

The analysis of pulse photoplethysmography (PPG) signals using computerized techniques is a developing field in research. Various effective signal-processing tools have been presented for automatic disease detection systems. Sleep apnea is a syndrome that affects the respiratory system and it commonly occurs due to Oxygen desaturation while sleeping. This paper develops an automatic system to detect sleep apnea from PPG signals. The detection of this syndrome is very important and many approaches were presented to improve the performance. The proposed method improves the classification accuracy through the enhancement of the feature extraction method and using the optimized classifier. As a feature extraction process, Hilbert Huang Transform (HHT) with extrema selection reformed (ESR) Empirical mode decomposition (EMD) is presented in this work. The development of the ESR-EMD system provides a better decomposition of signals and makes feature extraction effective. In addition, the computation time process is reduced as the interpolation is done using the more significant extrema points. Afterward, the feature selection is implemented using fisher discriminant analysis (FDA). An improved CNN classifier with a circular adaptive search butterfly optimization algorithm (CASBOA) is presented for classification. The optimum results obtained using BOA can be increased by employing an adaptive circular search function. This approach can increase the accuracy of the classifier and reduce computational time. The proposed approach is validated in MATLAB with a dataset and the performance metrics are compared with the conventional approaches.