A Novel Clustering Approach: Simple Swarm Clustering

Clustering categorizes data into meaningful groups without any prior knowledge. This paper presents a novel swarm-base clustering algorithm inspired from flock movement. Many algorithms solve the problem by optimizing a cost function but ours clusters data by applying one rule on data agent movements. We demonstrated that not only this simple rule is sufficient but completely effective in accurately dividing the data into natural clusters. It is a good model of how simply nature solves complex problems. Unlike some algorithms, this one does not need number of desired cluster in advance and discovers it by itself correctly. Eight data sets were used to compare the algorithm with five well-known algorithms. K-means and k-harmonic fail to find none-Gaussian clusters and two other swarm-base algorithms suffer severely from performance but our algorithm works successfully in both cases. The result confirms the superiority of our method.