An Evaluation on Competitive and Cooperative Evolutionary Algorithms for Data Clustering

Data clustering methods are important tools for exploratory data analysis in many real world applications, such as data mining, image understanding, text analysis, engineering, medicine, and so on. Partitional clustering models are the most popular clustering methods, but these approaches suffer from some limitations, like the sensibility to algorithm initialization and the lack of mechanisms to help them escaping from local minima points. Evolutionary Algorithms (EAs) are global optimization meta-heuristics known for their capabilities to find optimal solutions even when dealing with hard and complex problems. Although many EAs are based on competitive behavior among individuals, its is known that cooperation may lead to better solutions then sheer competition. In this work, we perform a comparative analysis among four state-of-the-art EAs (Genetic Algorithm, Differential Evolution, Particle Swarm Optimization and Group Search Optimization), implemented in both competitive and cooperative frameworks, in the context of data clustering problem. Experiments are executed using eleven real world benchmark datasets as the testing bed, so we could access whether competitive or cooperative behaviors would prevail. The experimental results showed that cooperative algorithms are able to find better solutions, in average, when dealing with clustering problems, than their corresponding competitive approaches, and such models also require less storage memory to keep their population in comparison to competitive methods.