Improved Accuracy Rates of a Prototype Based Classifier Using Evolutionary Computation

Prototype based classifiers allow to determine the class of a new example based on a reduced set of prototypes instead of using a large set of known samples. By doing this, the computational time gets substantially decreased as the initial set is replaced by a reduced one and hence the classification requires less computations to estimate nearest neighbours. In most simple classification problems the samples associated to each class are in general gathered in a particular region of the euclidean space defined by their characteristic features. In these particular problems prototype classifiers reach their best performance. Unfortunately, not all classification problems have their samples distributed in this way and therefore improvements are needed in order to reach acceptable classification accuracy rates. This work proposes a nearest prototype classifier that uses evolutionary computation techniques to increase the classification accuracy. A genetic algorithm was used to evolve the spatial location of each prototype resulting in a better distribution of prototypes which are able to obtain larger classification accuracy rates.