摘要
目前大多数客户分类研究常采用单一的分类模型。引入多分类器融合方法,并将其与自组织数据挖掘理论(SODM)相结合,提出了基于SODM的选择性融合算法(SBSF)。SBSF从全部基分类器的分类结果组成的初始模型集出发,由上一层模型两两组合产生新的待选模型,用最小二乘法来估计融合权重,而用外准则来评价和筛选中间候选模型,直到根据终止法则找到最优复杂度的融合模型。在15个UCI数据集上的实验结果显示,与单一的分类模型以及常用的多数投票法、贝叶斯方法、遗传算法等5种融合方法相比,SBSF具有更高的分类精度。进一步地,在信用卡数据集"german"上的客户分类实验表明,SBSF能自适应地从基分类器池中选择一个适当的基分类器子集进行融合,从而提高客户分类的精度。
At present,single model is adopted usually for customer classification.In this paper,multiple classifiers fusion is introduced and combined with self-organizing data mining（SODM）.Further,a SODM based selective fusion（SBSF） algorithm is presented.SBSF starts from the initial model set containing the classification results of all base classifiers,generates new candidate models in every layer through combining two of the previously selected models,utilizes least-squares to estimate the fusion weights,and uses external criterion to evaluate and select the candidate models,until we get the optima complexity fusion model by termination rule.The experimental results in 15 UCI data sets show that SBSF has higher classification accuracy than single classification model and five commonly used fusion methods,such as majority voting,Bayesian method,genetic algorithm.Further,the customer classification experiments on＂german＂data set fully display that SBSF can adaptively select an appropriate subset from the base classifiers pool to fuse,which can improve the classification accuracy.
出处
《管理工程学报》
CSSCI
北大核心
2010年第4期71-77,共7页
Journal of Industrial Engineering and Engineering Management
基金
国家自然科学基金资助项目(70771067)
