Electrocardiography Classification Based on Revised Locally Linear Embedding Algorithm and Kernel-Based Fuzzy C-Means Clustering

This paper presents an electrocardiographic signal classification method. This method first uses a revised locally linear embedding (LLE) algorithm to perform dimension reduction to the Electrocardiography (ECG) data. This manifold distance measurement-based LLE was proposed to solve the defect of conventional LLE that, due to the use of Euclidean distance, it could not properly measure the distance between high-dimensional samples. Using this revised LLE algorithm for the dimension reduction of data, more original data information could be retained, and the features of high-dimensional ECG data could be more effectively extracted, thereby improving the classification accuracy. The method then adopts kernel-based fuzzy C-means clustering algorithm to perform ECG signal classification. Classification tests on four common types of ECG signals-from the MIT-BIN database showed that the proposed method reached an overall accuracy of 99%.