Non-parametric Nearest Neighbor Classification Based on Global Variance Difference

Abstract As technology improves, how to extract information from vast datasets is becoming more urgent. well known, k-nearest neighbor classifiers are simple implement and conceptually implement. It not without its shortcomings, however, as follows: (1) there still a sensitivity the choice of k -values even when representative attributes considered in each class; (2) some cases, proximity between test samples nearest cannot be reflected accurately due measurements, etc. Here, we propose non-parametric classification method based on global variance differences. First, difference calculated before after adding sample subject, then divided by tested, resulting quotient serves objective function. In final step, tested classified into class with smallest discuss theoretical aspects this Using Lagrange method, it can shown that function optimal centers averaged. Twelve real University California, Irvine used compare proposed algorithm competitors such Local mean pseudo-nearest algorithm. According comprehensive experimental study, average accuracy 12 high 86.27 $$\%$$ % , which far higher than other algorithms. The findings verify produces results dependable existing