基于RRF的信息数据集的特征选择

Feature Selection of Information Data sets Based on RRF

摘要:

信息数据集中存在的冗余、不相关甚至是噪音特征会严重影响管理决策的质量和效率,笔者整合RS理论和ReliefF算法,提出了一种“二次过滤”的RRF组合决策方法。首先采用ReliefF算法计算出各个特征权重,过滤掉噪音特征,之后再用启发式的RS方法对原有特征集进行特征约简,过滤掉冗余特征。RRF组合决策方法集合了两个算法的优点,能够有效减少计算的工作量和复杂度,避免RS约简过多以及约简中存在噪音特征等问题,从而起到提高决策质量和效率的作用。最后,文章用一个真实的医疗信息数据集对算法进行了测试,并与两个常见的特征选择算法进行了对比,证明了本算法在辅助临床决策中的作用。

Abstract:

【】Redundant, irrelevant and even noise features may severely influence classification accuracy of data sets. A novel secondary filtration RRF group decision algorithm combining RS with ReliefF is proposed in this paper. RRF algorithm takes advantage of the two algorithms and can remove redundant attributes effectively, so the number of reducts is reduced and quality of reducts is improved. At last, the RRF algorithm is tested by one real medical data set and two common feature selection methods are used to compare with it. The experimental results show the effectiveness of RRF algorithm in clinical decision.