属性约简的两阶段遗传算法

属性约简是数据挖掘的一个重要研究内容. 为了解决具有多种属性类型的决策表约简问题,在粗集和二元关系聚合理论的基础上,利用属性重要性作为评价标准,提出了一种两阶段遗传约简算法. 算法的第一阶段是为了找出尽可能多的约简,第二阶段力求寻找最小约简. 根据算法每个阶段的目标设计了编码方案、种群规模、适应度函数、终止条件、选择、变异和修正操作. 实验表明,与标准遗传算法相比,两阶段算法在计算最小约简时更为准确和稳定.

Two-phase genetic algorithm for attributes reduction

Attribute reduction is an important aspect in data mining. In order to solve the reduction problems in a decision table with multiple attribute types, based on the theory of rough sets and binary relation aggregation, a two-phase genetic reduction algorithm is proposed using the attribute importance as evaluation criteria. The first stage of the algorithm is to find the reducts as much as possible, and the second stage is to find the minimal reduct. According to the goal of each stage, the coding scheme, the size of population, fitness function, termination condition, selection, mutation and correct operation of the two-phase genetic algorithm are designed. Experiments show that, compared with the standard genetic algorithm, the two-phase algorithm is more accurate and stable in the calculation of reduct.