【摘要】 针对基于熵的特征加权算法忽略了数据集内在特性对特征重要性的影响，导致特征选择效果不佳的问题，提出一种改进的基于熵的特征加权算法，根据信息熵计算特征维度的重要性权重，通过引入交叉验证实现不同数据集的阈值学习，确定用于度量特征重要性的最佳阈值参数，并基于该阈值对数据集进行特征选择。在微阵列数据集上的数值实验结果表明：相比于原算法，所提算法能够减少更多的维度，且特征子集用于分类得到的准确率与原算法基本持平甚至有所提高，说明改进的算法是可行和有效的。更多还原

【Abstract】 Aiming at the problem that the entropy-based feature weighting algorithm ignores the influence of the intrinsic characteristics of the dataset on the importance of features, which leads to poor feature selection, an improved entropy-based feature weighting algorithm is proposed, which calculates the importance weights of feature dimensions according to the information entropy, achieves threshold learning of different datasets by introducing cross-validation, so as to determines the optimal threshold parameter used to measure the importance of the features, and performs feature selection of the dataset based on this threshold. The numerical experimental results on the microarray dataset show that the proposed algorithm is able to reduce more dimensions than the original algorithm, and the accuracy of the feature subset used for classification is basically the same as the original algorithm or even better than that, which indicates that the improved algorithm is feasible and effective.更多还原