Comparison Of Various Kernels Of Support Vector Machine
Journal Title: International Journal for Research in Applied Science and Engineering Technology (IJRASET) - Year 2015, Vol 3, Issue 7
Abstract
As we know, classification plays an important role in every field. Support vector machine is the popular algorithm for classification and prediction. For classification and prediction by support vector machine, LIBSVM is being used as a tool. Support vector machine classifies the data points using straight line. Some datasets are impossible to separate by straight line. To cope with this problem kernel function is used. The central idea of kernel function is to project points up in a higher dimensional space hoping that separability of data would improve. There are various kernels in the LIBSVM package. In this paper, Support Vector Machine (SVM) is evaluated as classifier with four different kernels namely linear kernel, polynomial kernel, radial basis function kernel and sigmoid kernel. Several datasets are being experimented to find out the performance of various kernels of support vector machines. Based on the best performance result, linear kernel is capable of classifying datasets accurately with the average accuracy 88.20 % of correct classification and faster with 4.078 sec of prediction time. Radial basis function Kernel is capable of taking less training time compared to other kernels that is 4.92675 sec.
Authors and Affiliations
Supriya Pahwa, Deepak Sinwar
Keywords
Related Articles
- EP ID EP21134
- DOI -
- Views 253
- Downloads 8
Gas Level Alert and Automatic Cut-Off in a Stove
this paper proposes a technique which lets us to completely eliminate any mishappenings that generally occur with the leakage of gas in a daily base Gas stove system. The system can be operated in a way as to check whet...
A Clustering Based Efficient Intrusion Detecting In Multitier Dynamic Web Applications
We present DoubleGuard, a system used to detect attacks in multitiered web services and classify through Hierarchal clustering Algorithm. Our approach can create normality models of isolated user sessions that include b...
Synthesis of POA and study of its conducting and gas sensing properties
In this work, the poly (o-anisidine) (POA) powder was synthesized by chemical synthesis route. The resulting POA powder was characterized by UV-Visible absorption spectroscopy. A pellet of synthesized powder was prepare...
Fault Tolerant Protocol for UDS Flash Programming through Controller Area Network (CAN)
This paper proposes a fault tolerant protocol for CAN (Control Area Network) flash programming. Recently, the need of CAN flash programming is rapidly increasing because the number of installed ECU (Electronic Control...
A Hybrid Solar Tracking System with Photovoltaic Cells and Thermoelectric Module
Conventional sources like fossil fuels were used in early stages to satisfy the energy demands. Nowadays these are slowly replaced by renewable sources like photovoltaic sources. Solar PV is starting to play a substanti...