基于互信息和疑义度相结合的知识约简方法

A Method of Knowledge Reduction Based on Combination of Mutual Information and Doubtful Measure

提出一种基于互信息和疑义度相结合的知识约简方法.遵循修正的互信息准则,发展了一种类似于正交化特性的启发式算法,从决策系统中找出属性集的约简;该方法采用可增可删的双向回归算法,克服了目前前向选择或后向删除的知识约简方法中存在的属性相互依赖或依赖于决策类别的缺点,可保证分类精度不变的情况下,得到更为简化的决策属性集.最后,通过一个简单实例的仿真分析过程验证了文中所提方法的有效性.

A method of knowledge reduction based on the combination of mutual information and doubtful measure is proposed in this paper.On the basis of the modified mutual information criterion,a heuristic algorithm like the orthogonal property is developed to find the best reduction of the attribute set form the decision systems.By the addible and erasable stepwiseregression algorithm,this approach can overcome the shortcomings of the interdependence in the attribute set or the dependence on the decision classes existing in the recent knowledge reduction methods and can obtain simpler decision attributes with the univaried classification accuracy.Finally,a simple example is included to demonstrate the effectiveness of the algorithm.