利用粒分辩关系构建决策树的算法探讨

Study on algorithm of constructing decision tree using granulation relation

针对信息增益算法以信息量的大小确定首选属性和基于粗集算法以核属性为首选属性构建决策树的不准确问题,以粒理论为基础,将属性按其自然取值划分为若干基本粒,以条件粒的长度(粒分辩量)和该粒对决策粒分辩关系(分辩类别)为依据确定划分属性,采用简洁的算式解决了多属性的择优难题。理论和实例分析的结果表明,该算法具有建树精准简洁有效以及时空复杂度低的特点。

Aiming at the problem of inaccurate constructing a decision tree based on the information gain algorithm to determine the preferred attributes and an algorithm of the rough set using core attributes as the preferred attri- butes, in granulation theory, according to its nature value the attribute is divided into several basic granulation. Through conditional length （the volume of granulation resolution） and the resolution of granulation relativing to de- cision granulation（ the class of resolution）, the attribute is divided. Multi-attribute selection problem is solved by us- ing the simple formula. Theory and example analysis results show that the algorithm achieves precision and formula is simple effective and the temporal and spatial complexity is low.