Decision Boundaries and Classification Performance Of SVM And KNN Classifiers For 2-Dimensional Dataset

Hussain, Aini (2009) Decision Boundaries and Classification Performance Of SVM And KNN Classifiers For 2-Dimensional Dataset. Industrial Electronic Seminar.

Preview

PDF Download (583Kb) | Preview

Abstract

Support Vector Machines (SVM) and K-Nearest Neighborhood (k-NN) are two most popular classifiers in machine learning. In this paper, we intend to study the generalization performance of the two classifiers by visualizing the decision boundary of each classifier when subjected to a two-dimensional (2-D) dataset. Four different sets of database comprising of 2-D datasets namely the eigenpostures of human (EPHuman), the breast cancer (BCancer), the Swiss roll (SRoll) and Twinpeaks (Tpeaks) were used in this study. Results obtained confirmed SVM classifier superb generalization performance since it contributed the lower classification error rate when compared to the k-NN classifier during the training for binary classification of all 2-D datasets. This is evident and can be clearly visualized through the plots depicting the decision boundaries of the binary classification task.

Actions (login required)