Visualization of hyperplanes for SVM classification

Abstract

The 'Hyperplane' is the decision boundary in feature space that separates two classes with the greatest margin. This study aims to visualize SVM hyperplanes between multiple classes in a 3D feature space. This Visual Data Mining (VDM) tool is developed for four reasons: 1) to improve a user's understanding of the SVM classifier; 2) to visually assess the potential overlap of training pixels in feature space; 3) to assess the accuracy with which hyperplanes based on an SVM classifier can separate classes; 4) to explore uncertainty related to pixels that cross the hyperplane. This paper argues that VDM is an important tool for visual exploration of the data to improve insight into the classification algorithm and identify sources uncertainty.