Dissertation > Excellent graduate degree dissertation topics show

Holter waveforms clustering strategy effectiveness analysis

Author: MouShanLing
Tutor: ZhengGang
School: Tianjin University of Technology
Course: Applied Computer Technology
Keywords: Holter Waveform Clustering validity Internal evaluation methods Relative evaluation method FOM (Figure of Merit) Hausdorff distance
CLC: TP311.13
Type: Master's thesis
Year: 2011
Downloads: 40
Quote: 0
Read: Download Dissertation

Abstract

Clustering is unsupervised learning algorithm that data set in accordance with a clustering criterion is divided into different classes, the goal is to make as much as possible within the class of objects similar to the distance between classes as much as possible stay away from. The results of the cluster analysis, clustering algorithms are appropriate for a given data set, the resulting data set clustering results can reflect the inherent structure, which requires the clustering results were evaluated. Evaluate the effectiveness of traditional clustering methods are mostly for low-dimensional data clustering results, and achieved good results, but Holter waveforms for high-dimensional data, due to the presence of characteristic curve, the traditional effectiveness evaluation method there are some limitations. Based on the background, and Holter waveforms biological effectiveness evaluation methods to study the effectiveness of clustering results Holter waveform analysis, proposed for the ECG waveform analysis of the effect of clustering methods and internal evaluation relative evaluation method. This paper analyzes the characteristics of ambulatory ECG waveform, an improved method of FOM (Figure of Merit) ECG waveform clustering results for internal evaluation. FOM internal evaluation method is the classic method, but this method is reflected in the Euclidean distance differences within the class, is not suitable for ambulatory ECG waveform clustering results were evaluated. Minimal Hausdorff distance is a great distance, without creating one relationship between the points, just two points calculate the degree of similarity between sets. In this paper, FOM method, based on the ECG waveform segments by a weighted sum of Hausdorff distance calculation, we propose a method for improving the effectiveness evaluation, through the MIT-BIH ECG data in a standard database for the experimental results show that the FOM conventional method, the improved method can Holter waveforms effective evaluation clustering results. Clustering effect on the relative evaluation of a difficulty is to determine the optimal number of clusters category. This paper presents an approach based on the concept of gravity relative evaluation method, from the within-class compactness and inter-class separability perspective construct validity function, in order to determine the optimal number of clusters category. Through the MIT-BIH database of ECG data on the experiments indicate that the traditional classical methods SD (Scat-Dis) indicators and DB (Davies-Bouldin) compared the effect evaluation of the index, the proposed method has a better evaluation of results. Finally, the effectiveness of the proposed method and the evaluation of internal evaluation Holter waveforms relative evaluation method is applied to the evaluation of clustering results. By MIT-BIH standard database on the results of evaluation of different clustering algorithms, experimental results show the effectiveness of the proposed evaluation method can guide users to select the appropriate data set clustering algorithm, and can get the best clustering number of categories.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer software > Program design,software engineering > Programming > Database theory and systems
© 2012 www.DissertationTopic.Net Mobile