Finding representative patterns with ordered projections

Abstract

This paper presents a new approach to finding representative patterns for dataset editing. The algorithm patterns by ordered projections (POP), has some interesting characteristics: important reduction of the number of instances from the dataset; lower computational cost (Θ(mn log n)) with respect to other typical algorithms due to the absence of distance calculations; conservation of the decision boundaries, especially from the point of view of the application of axis-parallel classifiers. POP works well in practice with both continuous and discrete attributes. The performance of POP is analysed in two ways: percentage of reduction and classification. POP has been compared to IB2, ENN and SHRINK concerning the percentage of reduction and the computational cost. In addition, we have analysed the accuracy of k-NN and C4.5 after applying the reduction techniques. An extensive empirical study using datasets with continuous and discrete attributes from the UCI repository shows that POP is a valuable preprocessing method for the later application of any axis-parallel learning algorithm.