新能车属性离散化的分辨矩阵和信息增益算法

针对新能源智能车监控数据中包含过多的连续属性,提出了一种基于分辨矩阵和信息增益率的有监督离散化算法,从而降低连续属性的取值精度,使得新能源智能车后续的分类模型建立更具泛化能力。该算法在保证分类效果的前提下,获得尽可能少的结果断点,主要从3个方面对传统的离散化算法进行优化,一是根据决策表的条件属性与决策属性构建候选断点分辨矩阵,通过分辨矩阵判断相邻属性取值之间是否有可能的断点;二是用信息增益率来优化结果断点的选取;三是通过设定停止阈值解决了传统算法因停止条件过于严格导致算法选取过多的结果断点、离散化效果一般的问题。实验结果表明,改进的算法能够有效减少断点数量,大幅提高计算效率,并获得与经典算法相近的离散结果。

Candidate Cuts Matrix and Information Gain Algorithm for Discretization of New Energy Vehicle Attributes

In order to solve the problem that the monitoring data of new energy intelligent vehicle contains too many continuous attributes,a supervised discretization algorithm based on candidate cuts matrix (CCM) and information gain rate is proposed.Thus,the accuracy of continuous attributes is reduced,which makes the subsequent classification model of new energy intelligent vehicle more generalized.The algorithm obtains result breakpoints as few as possible on the premise of ensuring the classification effect.It optimizes the traditional discretization algorithm from three aspects.One is to build the CCM according to the condition attributes and decision attributes of decision table,and judge whether there is a possible breakpoint between adjacent attribute values through CCM.Another is to optimize the selection of the result breakpoint by the information gain rate.The third is to set the stop threshold to solve the problem that the traditional algorithm chooses too many result breakpoints and the general effect of discretization because of too strict stop conditions.The experimental results show that the improved algorithm can effectively reduce the number of breakpoints,greatly improve the calculation efficiency,and obtain the similar discrete results with the classical ones.