Generation of Alternative Clusterings Using Multi-objective Particle Swarm Optimization

Unsupervised nature of clustering algorithms makes it hard to discover the complex hidden patterns in the dataset. Conventional clustering methods discover only single clustering while complex patterns may have more than one pattern. These multiple patterns or alternative clustering can be valuable and interesting from the point of view of knowledge discovery. Alternative clustering offers various use cases from finance to environmental science and psychology to climatology. A multiple objective-based particle swarm optimization (PSO) approach is introduced for generating alternative clusterings. In multiple objectives, two objectives are being used: cluster quality within-cluster and dissimilarity between clusters. These objectives formulate minimization and maximization type of objective functions. Particle swarm optimization optimizes these objective functions efficiently and finds a wide scatter of solutions with good convergence to true Pareto optimal solutions. The obtained experimental outcomes indicate that the proposed approach is able to generate alternative clustering with a good amount of accuracy.