用数据挖掘技术实现多因素实验设计

Data Mining Techniques Used in Multifactor Experimental Design

目的本文介绍如何用数据挖掘技术实现同水平和混合水平多因素实验设计的方法,即独立设计;总结和讨论了独立设计的特点。方法以同水平或混合水平多因素析因设计矩阵为母体,不增加任何限定条件,仅考察各实验点对其余实验点的影响,采用“主成分分析、聚类分析和规则归纳”等数据挖掘技术相结合的方法,先将全部实验点快速分类,然后,再进行聚类,可将此矩阵分解成一系列彼此互不重叠的独立设计矩阵。结果不仅发现用数据挖掘技术实现多因素实验设计是可行的,而且,发现了除正交设计、均匀设计以外的一些可用于多因素实验设计的特殊设计。结论独立设计是一类涵盖面很宽的多因素实验设计方法,它不仅集“析因设计、分式析因设计、正交设计、均匀设计”于一身,还包含有突出“中间水平”或“极端水平”的特殊设计。与正交设计和均匀设计相比较,独立设计的适用面更宽、灵活性更大,具有极大的理论研究价值和广阔的实际应用前景。

Objective To illustrate how to realize multifactor experimental designs with same levels or mixed levels via the data mining techniques. Methods By taking the multifactor factorial design matrices as the basic design matrices, with no additional restrictions, the effects of each experimental point relative to other points were taken into consideration and combined data mining techniques were adopted to decompose the factorial design matrix into the desired independent design matrix. Results Not on- ly did we discover that it was feasible to realize multifactor experimental design, which can be called independent design, with the help of data mining techniques, hut also discovered some special designs which can be used in multifactor experimental designs in addition to the commonly used orthogonal and uniform design. Conclusion Independent design is a type of multifactor design of wide apphcahility, which is not only a quintessential representation of multifactor factorial design, fractional factorial design, orthogonal design and uniform design, hut also contains some special designs that give prominence to the ＂middle level＂ and ＂extremal levels＂ of some factors.