论文部分内容阅读
基于人工神经网络的特征选择算法一般可以看作是剪枝算法的一个特例:通过剪枝输入节点,计算网络输出对该输入节点对应特征的敏感性。但这些方法往往要求首先对数据做归一化的工作,这可能会改变原数据具备的对分类很重要的某些性质,神经模糊网络是具有自学习能力的模糊推理系统,本文将其与基于隶属度空间的剪枝技术结合起来提出新的特征选择算法,其特点是隶属度函数是自适应学习的,且学习过程在特征选择之前完成,分别对自然数据和人工数据进行实验,并与其它方法相比,结果证明该算法是有效的。
The feature selection algorithm based on artificial neural network can generally be regarded as a special case of the pruning algorithm: by pruning the input node, the sensitivity of the network output to the corresponding feature of the input node is calculated. However, these methods often require the data to be normalized first, which may change the original data possesses some properties that are important for classification. The neuro-fuzzy network is a fuzzy inference system with self-learning capability. Membership pruning space pruning technology combine to propose a new feature selection algorithm, which is characterized by membership function is adaptive learning, and the learning process is completed before the feature selection, respectively, the natural data and artificial data experiments, and other Compared with the method, the result proves that the algorithm is effective.