Similarity-based data reduction techniques

The k-nearest neighbours (kNN) is a simple but effective method for classification. Its major drawbacks are (1) low efficiency, and (2) dependency on the selection of a "good value" for k. In this paper, we propose a novel similarity-based data reduction method (SBModel) together with three variants aimed at overcoming these shortcomings. Our method constructs a similarity-based model for the data, which replaces the data to serve as the basis of classification. The value of k is automatically determined, is varied in terms of local data distribution, and is optimal in terms of classification accuracy. The construction of the model significantly reduces the amount of data needed for classification, thus making classification faster. Experiments conducted on some public data sets show that SBModel and its variants compare well with C5.0, kNN, wkNN, and other data reduction methods in both efficiency and effectiveness.