Comparison on Discriminant Analysis Using Logistic Regression and KNN Algorithm Based on Multilinear Principal Component Analysis for Study of Cardiology Ultrasound in Left Ventricle
| Autor: | Meng-Huan Hong, 洪夢寰 |
|---|---|
| Rok vydání: | 2013 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 101 Many studies in the literature pointed out that systolic and diastolic function of the heart is poor for patients with certain heart diseases relative to normal people. Here a two-dimensional echocardiographic image with gray-scale values is used as the two-dimensional response data to examine the heart functional risk. We make use of the differences between the systolic and diastolic in left ventricle of the heart images to analyze heart function and see if it is normal or abnormal. The objective of this study is to identify the important areas in left ventricle, as well as trying to use the information provided by the data to classify the people into two categories as with normal heart function or not. According to Yang (2012) and Chen (2012), in the analysis of cardiology ultrasound imaging data in left ventricle, while using principal component analysis (PCA in short) method, it will yield larger variations in the high-dimensional covariance matrix estimate with small sample size. Therefore, this study uses multilinear principal component analysis (MPCA in short) developed by Hung, H. et al. (2012) to extract important locations with larger variations which can better explained the variabilities. Later the original data is transformed into fewer variables for further investigations. Compared with the conventional PCA, the MPCA seems to be able to retain the data structure better and a better explained proportion of total variance. After extracting important variables for classification, logistic regression model and $k$-nearest neighbors algorithm as used to investigate the accuracy of the classification results. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|