模糊积分分类器中的自适应模糊测度

与其他分类器相比较,模糊积分分类器具有可以表示特征属性间交互作用的特性.确定合适的模糊测度是其关键因素之一.模糊测度的确定方法主要有2种:专家给定和从历史数据学习获得.由于模糊测度自身的复杂性,模糊测度主要是从数据中学习得到.为了能够更好地利用特征属性在样例空间体现出的局部特征,提出了一种用人工神经网络实现自适应模糊测度的方法.使得模糊测度可以随着输入样例的不同而变化,及时反映出在对样例进行分类过程中各特征属性的重要性和属性间的交互作用的不同,从而提高分类性能.实验证明该方法有效,可行.

Self-adaptive fuzzy measure for fuzzy integrals classifiers

By comparison with other classifiers,fuzzy integrals classifiers are peculiar in that it can express the interaction among features of examples.It is a key to determine appropriate fuzzy measures in fuzzy integrals classifiers.There are two main methods of determining fuzzy measures: specified by domain experts and learning from history data.Because of the complexity of fuzzy measures,fuzzy measures are determined from the history data mainly.To make good use of the local character of interaction among features in examples space,this paper proposes a method to achieve self-adaptive fuzzy measures based on neural networks.The fuzzy measure could change with different examples.The different importance of features and the interaction among features can be reflected in classification in time.The classification performance will be improved.Our experiment results show that this method is effective and feasible.