论文部分内容阅读
神经网络的输入属性选择一直是一个比较困难的问题.由于神经网络反复训练的时间复杂度,Wrap-per方法是不适用的,而单纯使用Filter方法也难以获得很好的分类精度.文中提出了一种两阶段的神经网络属性选择方法,以综合Filter和Wrapper两类方法的优势.该方法首先采用基于不一致率的遗传算法GFSIC来删除属性集合中的无关属性,然后采用基于敏感性度量的属性选择算法SBFCV来删除冗余和无用的属性.研究和实验结果表明,该方法可以有效地删除原始数据中的无关和冗余属性,增强神经网络的泛化能力.
It is always a difficult problem to choose the input attribute of neural network.Wrap-per method is not suitable due to the time complexity of neural network iterative training, and it is difficult to obtain good classification accuracy by using Filter method alone. A two-stage neural network attribute selection method, which combines the advantages of Filter and Wrapper, is proposed.Firstly, GFSIC based on the inconsistency-based genetic algorithm (GFSIC) is used to delete the irrelevant attributes in the attribute set, and then the attribute based on the sensitivity metric Select algorithm SBFCV to delete redundant and useless attributes.Research and experimental results show that this method can effectively remove irrelevant and redundant attributes in the original data and enhance the generalization ability of neural network.