A Novel Technique for Classification of Wrist Pulse for Health Monitoring using Sparse Classifier

Abstract: A wrist pulse system has been developed that can detect both normal and abnormal conditions in patients based on wrist pulse diagnosis. Pulse diagnosis are mainly done in three steps they are pulse preprocessing, feature extraction and classification. The acquired wrist pulse signal is passed through consecutive stages of denoising, baseline wander removal and period segmentation. The feature extraction is then done to extract time domain, frequency domain and wavelet features. Classification is then done for finding normal and abnormal conditions using SVM (Support Vector Machine) classifier. It is found that by using the SVM classifier, distributed features cannot be efficiently identified, classification accuracy is low and sub-classification cannot be done for abnormal condition as SVM supports only binary data. So SVM classifier is replaced by sparse classifier which has higher accuracy since it supports highly nonlinear data. T test is used in feature selection so that it needs low memory and less time consumption. Sub-classification has been done for the abnormal cases of Anemia, Arrhythmia and Wolff Parkinson White Syndrome. Keywords: Preprocessing, Feature Extraction, Sparse classifier, T test.