PREDICTION OF BREAST CANCER USING K-NEAREST NEIGHBOR CLASSIFIER WITH OPTIMAL K SELECTION

Abstract

In many scientific fields, machine learning methods have been widely used, but their use in medical literature is limited, partly due to technical difficulties. This research focuses on a machine learning method for predicting breast cancer with the aim of finding the optimum value of k in the k-Nearest neighbor classifier while dividing the data into different training and testing parts. The dataset used in this research work is “nki70” taken from the open.ml (https://www.openml.org/d/1147) machine learning repository having 77 features and 144 observations. The data set is a binary class problem with 48 observations of class 1 that represent breast cancer, and 96 observations of the other class 0, representing no breast cancer patients. The method considered in this research is the k-nearest neighbor algorithm. For selecting the best value of k, the performance metric, i.e., classification error rate, has been used. Results are given in the form of an average of 500 runs of the experiments on the different random training and testing sets. Furthermore, box plots of the results are also constructed. From the results of the analysis, it has been observed that the best value of k for which the k-NN classifier produced minimum error is k=1.