Super Box Cover And Binding KNN Classification Algorithm

The classification problem has become a very hot research in data mining. The purpose of classification is constructing a classifier according to the characteristics of the training dataset. With the use of the classifier, we can determine the class of the test sample or forecast the developmental trend. Among all the classification algorithms, KNN algorithm is widely used in data mining due to its simplicity and efficiency. However, it may need huge computation if KNN is directly used in the training samples.In this paper, classification algorithm based on field covering of multiple hyper-cubes and KNN is proposed to reduce the cost of computation. The base idea is that constructing a group of hyper-cubes to cover the training samples for every class in training dataset. A class of hyper-cubes has no intersection with other class of hyper-cubes. In testing phase, if the test sample falls within one of hyper cubes, the class of this sample is recognized. Otherwise, we can recognize the class of this sample according to KNN. The distance between the sample and the center of the hype-cube, and the distance between the sample and the surface of the hyper cube are all considered to calculate the similarity while we use KNN to identify the class of the testing sample.Finally we test the efficiency of this algorithm in four datasets. Comparing the result with other algorithms, we conclude that the algorithm base on hyper cubes and KNN is an effective classification method.