一种面向不平衡分类数据的核求解算法

An algorithm for computation of a core for unbalanced classification data

对基于差别矩阵的核求解方法而言,差别矩阵的规模是直接影响核求解效率的关键因素.为此,针对不平衡分类数据情况,提出一种基于多差别矩阵的核求解算法.该算法先按决策属性值划分对象集,进而建立任意两个不同对象集对应的差别矩阵,形成多差别矩阵,从而求出核.各差别矩阵因不平衡分类数据可有效降低其规模,提高核的求解效率.理论分析和实验结果表明算法是有效可行的.

For the method based on discernibility matrix for computing a core, reducing the size of discernibility matrix is the key for improving the performance of computation of a core, Therefore, an algorithm （AMDMC） based on multi-discernibility matrix is introduced to computation of a core for the case of unbalanced classification data. By the decision attribute＇s value, the all objects are partitioned into some subsets. For any two different subsets, a subdiscernibility matrix is created. Finally, the multi-discernibility matrix is obtained and a core is acquired. Each subdiscernibility matrix holds a small space because of unbalanced classification data, so the AMDMC algorithm is in high efficiency. Theoretical analysis and experiment results show the effectiveness of the algorithm.