GAKC: A New GA-Based k Clustering Algorithm
Xiaohong Li, Luo Min · 2009
Clustering is an important, hard and active topic in data analysis and pattern recognition. K clustering is a branch of data clustering where the number of clusters is know in advance. Recently, spectral clustering (SC) becomes one of the most popular and appealing k clustering methods because of its generality, efficiency and its rich theoretical foundation. But the final results obtained from SCs depend on spectral relaxation which may have no guarantee on the quality of the solution. In order to overcome the SCs' shortcoming, we propose an effective GAKC algorithm by using a genetic algorithm to search for the optimal cluster result of SCs. The algorithm uses group number coding chromosome, a new uniform crossover operator and exponential mutation rate. To verify the effectiveness of GAKC, a comparison among the experimental results of the proposed GAKC, a classical GA-based method by Ujjwal Malulik and the SC methods by SM and NJW on a real-life data set is presented. The conclusion comes that the proposed algorithm can gain much more accurate clustering result.