Using Heuristic Model to Improve the Efficiency of Support Vector Machines

Abstract

Support Vector Machines (SVM) have become increasingly popular tools for many data mining tasks. It can be used in classification, novelty detection, regression, and clustering. It has been successfully applied to a lot of applications about text categorization, handwritten character recognition, medical diagnosis, bioinformatics and database marketing. However, the application of SVM to large datasets is limited because of the high computational cost involved in solving quadratic programming problem arising in training. To solve this problem, this research tried to develop a heuristic model to reduce the computational and space cost. The model is composed of three parts. 1. Finding the principal attributes by PCA. 2. Error-tolerance constraints for lossy compression. 3. Replaced values computation and similar records deletion. Then we apply SVM on the compressed database. The experimental results have proved that the heuristic model will reduce the input features to save the memory and the computation time. And the accuracy is acceptable even improved.