基于粗糙集理论的两类离散化方法研究

连续属性离散化是数据预处理的关键步骤之一,在实际应用中往往通过高效的启发式算法来计算离散化结果.对基于辅助矩阵和信息熵的两类启发式离散化算法进行实验研究,分别选取每类算法中的5种典型方法,通过系列实验,对两类算法的性能进行对比研究,结果表明:辅助矩阵类算法具有相对较高的样本识别能力,但算法复杂度较高,运行时间更长,较适...

Study on two kinds of discretization methods based on rough set theory

The discretization of continuous attributes is one of the key steps of Data Pretreatment. In practical application, continuous information systems are usually discretized by efficient heuristic algorithms. Herein, two kinds of heuristic data discretization approaches, i.e. discretization methods respectively based on auxiliary matrixes and information entropy, are thoroughly studied by simulation experiments. Five typical algorithms of each kind are realized and their performances are comprehensively compared by a series of experiments. The experimental results suggest that auxiliary matrix based algorithms are with higher capability, but more complex and time consuming, thus is appropriate for small-scaled continuous systems; and the characteristics of information entropy based algorithms are on the contrary.