Performance comparison of two non-parametric classifiers for classification using geometric features

Abstract

This study aims to examine and compare the performances of Random Forest (RF) and k-Nearest Neighbor (k-NN) algorithms used for classification based on certain geometric features. For the purpose of the analysis, the Breast Cancer Wisconsin (BCW) public dataset is used. BCW dataset contains features like area, perimeter, radius, compactness, and symmetry computed from 357 benign, and 212 malignant breast images, respectively. Three different experiments related to the size of training and testing datasets for classification are conducted and different accuracy values are obtained. The best accuracy of 91.9% for RF and 91.3% for kNN, respectively, are reached when 30% of the entire dataset is used as testing dataset. For all experiments, the RF classifier outperformed the kNN.