一种基于粗糙集和信息熵理论的属性离散化算法

AN ATTRIBUTE DISCRETISATION ALGORITHM BASED ON ROUGH SET AND INFORMATION ENTROPY THEORY

在分析和研究C5算法中连续属性处理的必要性及C5算法中离散化方法的不足后,采用基于粗糙集理论-信息熵-可辨识矩阵的离散化的方法(RSIEDM)进行离散化。该方法利用粗糙集、信息熵和可辨识矩阵能更合理、更准确地对连续属性进行离散化,使创建的决策树具有更好的准确率。在优化雷电灾害统计和评估雷电灾害导致的损失应用中,该算法取得了较好的效果。

After analysing and studying the necessity of continuous attributes processing in C5 algorithm and the shortcomings of discretisation method in C5 algorithm, we propose to adopt a discretisation method based on rough set theory-information entropy-discernible matrix discretisation （RSIEDM）. By making use of the above three, the method can more reasonably and accurately discretise the continuous attributes, and enables the created decision tree to have better accuracy. Applying the method to optimising the statistics of lightning disaster and evaluating the losses of lightning disaster, the algorithm achieves desirable result.