Automatic detection of k with suitable seed values for classic k-means algorithm using DE

摘要

k-means algorithm, in spite of its computational efficiency and capacity for faster convergence has some serious drawbacks like its tendency to stick into local optima and the requirement of supplying number of cluster before execution. Our algorithm used Differential Evolution (DE) as preprocessor to overcome those bottlenecks. Experiments show that the improved version of clustering algorithm produces better results.