Clustering Problem with 0-1 Quadratic Programming

The most unsupervised methods of classification suffer from several performance problems, especially the class number, the initialization start points and the solution quality. In this work, we propose a new approach to estimate the class number and to select a set of centers that represent, fiddly, a set of given data. Our key idea consists to express the clustering problem as a bivalent quadratic optimization problem with linear constraints. The proposed model is based on three criterions: the number of centers, the density data and the dispersion of the chosen centers. To validate our proposed approach, we use a genetic algorithm to solve the mathematical model. Experimental results applied on IRIS Data, show that the proposed solution selects an adequate centers and leads to a reasonable class number.