A genetic algorithm based clustering approach for improving off-line handwritten digit classification

In this paper a new clustering technique for improving off-line handwritten digit recognition is introduced. Clustering design is approached as an optimization problem in which the objective function to be minimized is the cost function associated to the classification, that is here performed by the k-nearest neighbor (k-NN) classifier based on the Sokal and Michener dissimilarity measure. For this purpose, a genetic algorithm is used to determine the best cluster centers to reduce classification time, without suffering a great loss in accuracy. In addition, an effective strategy for generating the initial-population of the genetic algorithm is also presented. The experimental tests carried out using the MNIST database show the effectiveness of this method.