يك سيستم خوشه بندي تركيبي براساس الگوريتم تكاملي تفاضلي براي تعيين موثر مراكز خوشه هاي اوليه و روشهاي مختلف فاصله يابي دو به دو براي خوشه بندي نهايي

A Hybrid Clustering System Based on, (DE) Algorithm for Setting Efficient Initial States, and Diverse Pairwise Distances for Clustering

In recent decades, various clustering methods and systems have been proposed, but a lot of them had problem in selecting initial clusters centers. In this paper a new clustering system based on Differential Evolution algorithm (DE) for fixing initial clusters centers problem and different pairwise distances like (Euclidean, City-block and Chebyshev) for setting final clustering output, has been proposed. After choosing initial clusters centers by differential evolution algorithm, a kind of process to changing clusters centers will be applied, which is based on finding nth farthest samples from each cluster and then calculating the mean for each nth farthest samples for each cluster, and finally inverting the calculated mean for each nth farthest samples. With this approach, problem in choosing initial clusters will fixes. After choosing best initial clusters, the main process of finding and biasing clusters centers and allocating samples to closest cluster take place. This system has been validated with some of benchmark clustering datasets like Fisher-iris, Ionosphere and User-Knowledge-Modeling and compared with famous clustering methods like K-Means, Fuzzy C-means (FCM) and Gaussian Mixture model (GMM) and returned satisfactory results.

In recent decades, various clustering methods and systems have been proposed, but a lot of them had problem in selecting initial clusters centers. In this paper a new clustering system based on Differential Evolution algorithm (DE) for fixing initial clusters centers problem and different pairwise distances like (Euclidean, City-block and Chebyshev) for setting final clustering output, has been proposed. After choosing initial clusters centers by differential evolution algorithm, a kind of process to changing clusters centers will be applied, which is based on finding nth farthest samples from each cluster and then calculating the mean for each nth farthest samples for each cluster, and finally inverting the calculated mean for each nth farthest samples. With this approach, problem in choosing initial clusters will fixes. After choosing best initial clusters, the main process of finding and biasing clusters centers and allocating samples to closest cluster take place. This system has been validated with some of benchmark clustering datasets like Fisher-iris, Ionosphere and User-Knowledge-Modeling and compared with famous clustering methods like K-Means, Fuzzy C-means (FCM) and Gaussian Mixture model (GMM) and returned satisfactory results.