aiNet clustering algorithm based on objective evolution

作者：Guo, Jian-Hua^*; Deng, Fei-Qi; Yang, Hai-Dong

来源：Journal of South China University of Technology(Natural Science Edition), 2009, 37(4): 1-6.

• Artificial immune networks
• Clustering analysis
• Multi-objective planning
• New clustering algorithms
• Nonlinear clustering
• Objective evolution
• Parameter sensitivities
• Vaccination strategy

As the aiNet algorithm has no objective function and possesses a memory network with irregular and dynamic change, a new clustering algorithm of artificial immune network based on the objective evolution is proposed and is marked as OE-aiNet. In this algorithm, the compression and clustering based on artificial immune network is abstracted as a multi-objective planning problem, the objectives to which the memory network evolves is defined, and the quality of immunity learning is improved by adopting the vaccination strategy. Simulated results of kernel clustering and nonlinear clustering prove that (1) OE-aiNet is better than the existing aiNet algorithm in terms of clustering quality, compression quality and parameter sensitivity; (2) the average trace of class spread matrix of OE-aiNet, namely 4.1420, is less than that of aiNet (4.2575); (3) the compression ratio of OE-aiNet is 8.42% higher than that of aiNet; and (4) the clustering accuracy of OE-aiNet is not as sensitive to the compression threshold as that of aiNet.

• 出版日期2009-4
• 单位华南理工大学