Correlation Coefficient-based K-means Clustering for K-NN

Abstract

K-nearest neighbor algorithm is one of the most popular classifications in machine learning zone. However, as k-nearest neighbor is a lazy learning method, when a system bases on huge amount of history data, it faces processing performance degradation. Many researchers usually care about only classification accuracy, but the speed of estimation also play an essential role in real time prediction systems. For this issue, this research proposes correlation coefficientbased k-mean clustering for k-nearest neighbor aiming at upgrading the performance of k-nearest neighbor classification by improving processing time performance. For the experiments, we used the real data sets, Breast Cancer, Breast Tissue and Iris, from UCI machine learning repository. Moreover, the real traffic data collected from Ojana junction, Route 58, Okinawa, Japan, was also utilized to show the efficiency of this method. By using these datasets, we prove the better processing performance and prediction accuracy of the new approach by comparing the classical k-nearest neighbor with the new k-nearest neighbor.