基于局部分布的贝叶斯自适应共振理论增量聚类算法

Incremental clustering algorithm of Bayesian adaptive resonance theory based on local distribution

针对传统的贝叶斯增量聚类算法需要人为设置参数,且对分布不均衡数据聚类效果不佳的问题,提出一种基于局部分布的贝叶斯自适应共振理论增量聚类算法.首先,利用数据快照读取数据;然后,在无需设置参数的情况下,考虑类簇的局部分布情况,自适应地确定新数据的所属类别,并更新获胜类簇;最后,确定相邻快照中类簇的演化关系.不同数据集的仿真结果表明,所提出的算法在准确性和自适应性方面均有显著提高.

Traditional incremental clustering algorithm needs to be set parameters and cannot deal with the imbalance data.To solve the problem, the incremental clustering algorithm of bayesian adaptive resonance theory based on local distribution is proposed. Firstly, the new data are collected by data snapshots. Then, in current data snapshot, the new data are clustered into the winning cluster adaptively according to the local distribution of the clusters without predefined parameters. Then, the evolving relationships between the clusters in two neighboring data snapshots are determined.Finally, the simulation result shows that the proposed algorithm can improve the accuracy and the adaptability.