Research and application of extension theory-based k-nearest neighbors data-classification

In the field of data mining, the data classification is an important part of data analysis, which is used to determine the sample category and further extract information and knowledge for the decision-making. K nearest neighbors (KNN) is one kind of the classification methods. Although it can realize the classification without the prior parameter for the data processing, the classification accuracy is not high that the result is not ideal enough. Combining the extension theory and the characteristics of data classification, an extension K nearest neighbors (EKNN) is proposed, in which the matter-element model is used to describe the data in a triple way, the extension distance is applied to realize the calculation of data similarity, and the attribute reduction is introduced for the data-classification. Thought the experiments on three different UCI datasets, EKNN is apparently more effective and extensible than traditional KNN, which has a unified and clear data-description, effective data-classification process and higher classification accuracy.