Effect of Different Kernels on the Performance of an SVM Based Classification
Deepika Kancherla1, Jyostna Devi Bodapati2, Veeranjaneyulu N3
1Deepika Kancherla, Student, Department of CSE, Vignan’s University, Guntur (Andhra Pradesh), India.
2Jyostna Devi Bodapati, Asssistant Professor, Department of CSE, Vignan’s University, (Andhra Pradesh), India.
3Veeranjaneyulu N, Professor, Department of IT, Vignan’s University, (Andhra Pradesh), India.
Manuscript received on 12 February 2019 | Revised Manuscript received on 02 March 2019 | Manuscript Published on 08 June 2019 | PP: 1-6 | Volume-7 Issue-5S4, February 2019 | Retrieval Number: E10010275S419/19©BEIESP
Open Access | Editorial and Publishing Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: According to the literature Support Vector Machines (SVM) is one of the robust classification models which guarantees reasonable per-formance even with small training datasets. Though the deep learning models are able to produce the state of the art performance large volumes of training data is required to achieve that. SVMs are basically designed to be binary classifiers and can be extended to multiple classes that are very common in many real world applications. In this paper we are trying to prove that generalization ability of support vector machines (SVM’s) is good on difficult real world problems. We also try to analyze the effect of different features and different types of kernels on their performance. For the illustrations we have used different types of features like gist, HOG, histogram. In this work we show how the types of features extracted from the data can affect the performance of the classifier. The original version of SVMs is designed for linear classification tasks which can be applied to non-linear classification by projecting the data into a non-linear space using kernel trick. In this paper we even try to analyze the effect of kernels like linear, polynomial, Gaussian, sigmoidal and user defined kernels and how the type of kernel effect the performance of the support vector machine based classification task. Based on the studies we have conducted, it is observed that type of features and type of kernels used have a great impact on the performance of an SVM based classification task. Type of the features we can use is solely dependent on the problem on hand. On the other side impact of the kernel is dependent on the data set. Our Studies show that RBF kernel and histogram intersection kernel leads to better performance than others.
Keywords: Histogram Intersection Kernel; Kernel Trick; SVM; Types of Kernels; User-Defined Kernels.
Scope of the Article: Classification