Modeling Gestalt laws for classification

The k-nearest neighbors classifier is simple and often results in good classification performance on problems with unknown and non-normal distributions. However, its selected nearest neighbors on noisy, sparse, or imbalanced data are often inconsistent with our intuition and in turn leads to the worse performance. This paper applies Gestalt visual perceptual laws to design a new KNN classifie r. It applies the neighborhood relation between any two data points to construct the geometry shape of data and then applies the Gestalt laws on this shape to perform the classification. The conducted experiments on challenging benchmark real data validate the proposed approach.