基于流形学习降维的决策分析算法

为了有效地分析高维决策表，提出了基于流形学习降维的决策分析算法（decision analysis algorithm based on manifold learning，DAML）. 算法使用等距映射法（ISOMAP）对原始数据做降维处理，在得到的主坐标数据上进行决策分析. 根据核主成分分析法与ISOMAP方法的关系得到主成分与主坐标的转换关系式，并计算原始数据主成分. 提出了基于等价支持子集的决策算法用于计算主成分属性重要性、属性区分能力及等价支持子集. 在得到等价支持子集的基础上抽取决策规则，根据决策规则预测算法预测未知数据. 选取UCI数据库中标准分类数据集作为仿真实验样本，并对比C4.5决策树算法、K最近邻居算法（KNN）与提出的决策规则预测算法在Iris、Breast cancer、Wine、Spectf heart和Ionosphere数据集上的分类精度来验证算法的有效性.

Decision analysis algorithm based on manifold learning dimension reduction

A decision analysis algorithm based on manifold learning (DAML) is proposed for high-dimensional decision information system. Firstly, the dimension of the original data is reduced by using isometric mapping (ISOMAP) in the algorithm. Some decision rules are extracted from the low-dimensional embedding of manifold learning. The principal component of the original data is computed by the relationship of kernel principal component analysis and ISOMAP. Then, a support subset significant algorithm is proposed for decision analysis. The support subset significant algorithm is used to computing the attribution significant, discrimination power and support subset. The decision rules are extracted by the support subset. Finally, the Iris, Breast cancer, Wine, Spectf heart and Ionosphere on UCI benchmark are selected for simulation. The classification accuracy among C4.5, k-nearest neighbor (KNN) and DAML are compared on the 5 data sets to validate the effectiveness of the DAML.