Abstract
Researchers have proposed several Genetic Algorithm (GA) based crisp clustering algorithms. Rough clustering based on Genetic Algorithms, Kohonen Self-Organizing Maps, K-means algorithm are also reported in literature. Recently, researchers have combined GAs with iterative rough clustering algorithms such as K-means and K-Medoids. Use of GAs makes it possible to specify explicit optimization of cluster validity measures. However, it can result in additional computing time. In this paper we compare results obtained using K-means, GA K-means, rough K-means, GA rough K-means and GA rough K-medoid algorithms. We experimented with a synthetic data set, a real world data set, and a standard dataset using a total within cluster variation, average precision, and execution time required as the criteria for comparison.
Highlights
Grouping of large number of objects into a smaller number of manageable groups makes it easier to formulate planning strategies
The focus of this paper is to study the effectiveness of evolutionary rough clustering techniques
Rough clustering is more flexible than the conventional crisp clustering as it allows for an object to belong to more than one cluster with the help of boundary regions
Summary
Grouping of large number of objects into a smaller number of manageable groups makes it easier to formulate planning strategies. The K-means as well as the K-medoid clustering partition the input data set into k clusters to optimize an objective partitioning criterion, such as dissimilarity function based on distance. The objective partitioning criterion is not explicitly optimized Another drawback of K-means and K-medoid clustering is that it can fall in local optima. In this paper we discuss various aspects of crisp, rough set based, and evolutionary clustering algorithms. We discuss and present appropriate modifications required to the basic K-means and K-medoid algorithms so that these algorithms adapt to rough and evolutionary clustering. We discuss the formulation of a fitness function for GA based crisp and rough clustering .
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have