论文部分内容阅读
为了克服Relief选择前k个特征作为约简子集所存在的原始特征空间中的近邻在约简后的特征子空间中不一定还是近邻的问题,提出了一种在特征子空间中评价候选特征子集类别区分能力的方法,并结合最好优先特征搜索策略提出了一种新的特征子集选取方法.在12个UCI(加州大学欧文分校)数据集和1个老年痴呆实测数据集上,就约减能力对所提方法与其他3种经典特征选择方法进行了比较,并用决策树、逻辑回归模型详细比较了分类效果.实验结果表明:所提方法不仅能够选出特征数目较少的特征子集,而且特征子集的分类效果良好.
In order to overcome the problem that Relief selects the first k features as reductions and the neighbors in the original feature space of the reduced feature subspace are not necessarily or neighbors in the reduced feature subspace, a method of evaluating candidate features in the feature subspace is proposed A new feature subset selection method is proposed based on the best prioritized feature search strategy.On the basis of 12 UCI (Irvine, U.K.) datasets and 1 senile dementia data set, The proposed method is compared with the other three classical feature selection methods in terms of reduced abilities and the classification results are compared in detail by using the decision tree and the logistic regression model.The experimental results show that the proposed method can not only select features with fewer features Subset, and the classification of the feature subset works well.