Classification of photoplethysmography signals using features of sequential forward mother wavelet selection method

Photoplethysmography (PPG), a painless, simple, low-cost and non-invasive optical method used to detect blood volume changes in the vascular tissue bed, is one of the physiological data that is increasingly used in medicine and engineering. Especially the developments in pattern recognition and machine learning algorithms have increased the interest in applications based on these signals. By using PPG signals, methods such as heart and blood pressure disorders, stress level, person recognition and fatigue detection are suggested. These studies offer new feature extraction approaches by dealing mostly with two-class problems. Although the results obtained are relatively successful, they are still at performance levels that are open to improvement. In this thesis, it is proposed to classify PPG signals with high performance with sequential forward main wavelet selection method features. The validity and stability of the proposed method was applied to 3 separate PPG datasets. The features were obtained by calculating the standard deviation and average values of the wavelet transform coefficients obtained with the determined main wavelets. Then, the obtained features were classified using the k-nearest neighborhood method. The proposed method achieved average classification accuracies of 65.76%, 73.39%, and 86.54% for Data Set 1 (determination of mental workload level), Data Set 2 (detection of hypertension disease), and Data Set 3 (human activity monitoring), respectively. The obtained results showed that the proposed method has significant potential in classifying PPG signals.