A hybrid binary classifier: Using modified Logistic Regression for non-support vector elimination

This paper is a report on a new Hybrid Binary Classifier that aims to eliminate non-support vectors through a pre-processing stage and hence aims to reduce the storage and time requirements for the training phase of an SVM classifier without forgoing accuracy. The paper investigates the possibility of dividing the N-dimensional space into 3 sub-regions - one each for both labels and the third which holds the region of contention between the 2 labels. The new classifier is built using a modified form of Logistic Regression and SVM. Such a classifier is tested on a number of datasets and the findings are reported.