ABSTRACT
In this chapter, we describe our novel approach to enhancing the training process of SVM when dealing with large training datasets. It is based on a combination of SVM and clustering analysis. The idea is as follows: SVM computes the maximal margin separating data points; hence, only those patterns closest to the margin can affect the computations of that margin, whereas other points can be discarded
without affecting the final result. Those points lying close to the margin are called support vectors. We try to approximate these points by applying clustering analysis. The novelty of our approach lies in the algorithm that we have developed, called the Dynamically Growing Self-Organizing Tree (DGSOT). We will see that this algorithm facilitates hierarchy construction. In this chapter, we will describe DGSOT, and in Chapter 6, we will discuss how it is applied to intrusion detection.
