PCA Based Dimension Reduction of Feature Matrix to Train SVM for Balance Disorder Diagnosis

SERHAT, IKIZOÄžLU (2018) PCA Based Dimension Reduction of Feature Matrix to Train SVM for Balance Disorder Diagnosis. In: Eighth International Conference On Advances In Computing, Control And Networking - ACCN 2018, 23-24 June, 2018, Paris, France.

[img]
Preview
Text
20180712_063203.pdf - Published Version

Download (849kB) | Preview
Official URL: https://www.seekdl.org/conferences/paper/details/9...

Abstract

This study is mainly about the research to select the discriminative features for the machine learning algorithm to figure out the reason behind the problem of people who suffer from balance disorder. The foregoing step on this way has been determining the proper algorithm where we achieved the best performance with the Support Vector Machine (SVM) with Gaussian Kernel, the so-called Radial Basis Function (RBF). In our study, we first input the complete IMU-sensor based data set collected both from the healthy people and those suffering from vestibular system disorders to SVM-RBF. Next, we reduce the feature matrix using the Principle Component Analysis (PCA). Following this procedure, the machine is trained with the new data to recognize the effect of feature transformation on the accuracy of the learning method. We observed that PCA had satisfactory influence on the elimination of redundant features that it points to high correlation between some of the members of the starting feature matrix. The study will continue to cover more input vectors to PCA. Moreover, we plan sub-classification between various problems that lead to balance disorder. The situation with the current outputs of the study encourages to go further steps to achieve a significantly high performance for the machine learning algorithm with reasonable number of features.

Item Type: Conference or Workshop Item (Paper)
Uncontrolled Keywords: principle component analysis, machine laerning, support vector machines, vestibular system
Depositing User: Mr. John Steve
Date Deposited: 08 Mar 2019 14:50
Last Modified: 08 Mar 2019 14:50
URI: http://publications.theired.org/id/eprint/147

Actions (login required)

View Item View Item