Wrist pulse diagnosis using LDA

Abstract

The wrist pulse signal is a kind of important physiology signal which can be used to analyze a person's health status. This paper applies a linear discriminant analysis (LDA) to extract feature and used k-nearest neighbor (KNN) algorithm to distinguish the patients from health. In order to reduce the interference of noise, we first drew a series of pulse data of good quality from the original wrist pulse signal. We then reduced all high dimensional pulse signals to low dimensional feature vectors using LDA. Finally, we used a KNN algorithm to distinguish healthy persons from patients. The classification accuracy is over 83% in distinguishing healthy persons from patients with all kinds of diseases, and over 92% for single specific disease. The experimental results indicate that LDA is an efficient approach in telling healthy subjects from patients of specific diseases. © 2010 Springer-Verlag.