Affiliations: [a] Department of Computer Science and Engineering, The M.S. University of Baroda, Vadodara, Gujarat, India | [b] Computer Engineering Department, Charotar University of Science and Technology, Changa, Gujarat, India
Correspondence:
[*]
Corresponding author: Hetal Bhavsar, Department of Computer Science and Engineering, The M.S. University of Baroda, Vadodara, India. Tel.: +91 9879628487; E-mail:het_bhavsar@yahoo.co.in
Abstract: The Support Vector Machine (SVM) is a powerful technique for data
classification. For linearly separable data points, the SVM constructs an
optimal separating hyper-plane as a decision surface, to divide the data
points of different categories in the vector space. For the non-linearly
separable data points, the Kernel functions are used to extend the concept
of the optimal separating hyper-plane so that the data can be linearly
separable. The different kernel functions have different characteristics and
hence the performance of the SVM is highly influenced by the selection of
kernel functions. This paper presents the classification algorithm that uses
the SVM in the training phase and the Mahalanobis distance in the testing
phase, in order to design a classifier which has low impact of kernel
function on the classification accuracy, positively. The Mahalanobis
distance is used to replace the optimal separating hyper-plane as the
classification decision making function in the SVM. The proposed approach is
compared with Euclidean-SVM, which uses Euclidean distance function to
replace the optimal separating hyper-plane as the classification boundary.
It has also been evaluated against conventional SVM too. The experimental
results show that the accuracy of the EuDiC (Euclidean Distance towards the
Center of data) SVM classifier has a low impact on the implementation of
kernel functions. The EuDiC SVM also achieves the drastic reduction in the
classification time since it only depends on the mean of Support Vectors
(SVs) of each category for classification. To prove its effectiveness on
other types of data, the time series data have also been used. Due to robust
design of the EuDiC, it also performs well for time series data too.
Keywords: Classification, Euclidean distance, kernel function, Mahalanobis distance, optimal hyper-plane, support vector machine, support vector
