Particle swarm optimization for support vector clustering Separating hyper-plane of unlabeled data

The objective of this work is to design a new method to solve the problem of integrating the Vapnik theory, as regards support vector machines, in the field of clustering data. For this we turned to bio-inspired meta-heuristics. Bio-inspired approaches aim to develop models resolving a class of problems by drawing on patterns of behavior developed in ethology. For instance, the Particle Swarm Optimization (PSO) is one of the latest and widely used methods in this regard. Inspired by this paradigm we propose a new method for clustering. The proposed method PSvmC ensures the best separation of the unlabeled data sets into two groups. It aims specifically to explore the basic principles of SVM and to combine it with the meta-heuristic of particle swarm optimization to resolve the clustering problem. Indeed, it makes a contribution in the field of analysis of multivariate data. Obtained results present groups as homogeneous as possible. Indeed, the intra-class value is more efficient when comparing it to those obtained by Hierarchical clustering, Simple K-means and EM algorithms for different database of benchmark.