逻辑模型树算法性能分析与改进研究

The research on analysis and improvement of logistic model tree

逻辑模型树(LMT)算法是基于树归纳和逻辑回归的一种分类算法。为验证LMT算法的优势,利用3个UCI标准数据集建模,将LMT算法与其他决策树方法进行对比分析。针对LMT算法在建立逻辑回归模型时会导致较高的计算复杂性的问题,研究利用赤池信息量准则改进LMT算法,提升算法时间性能,避免模型过度拟合。在UCI标准数据集和烟叶综合质量评价数据中应用改进的LMT算法进行建模验证,结果表明,该改进方法在模型精度和召回率方面基本优于其他决策树方法,时间性能比改进前提升50%左右,能较好地评价烟叶综合质量。

Logistic Model Trees （LMT） algorithm is a classification algorithm which is based on tree induction and logistic regression. To verify the advantage of LMT, compare and analyze LMT with other decision tree methods on three UCI data sets. Because in logistic model trees, logistic regression models can lead to the high computational complexity. This issue can be addressed by using the AIC criterion to improve LMT. It can improve time performance of algorithm and prevent over fitting models. The modification of LMT is used on UCI data sets and tobacco comprehensive quality evaluation data. And the result demonstrates that this method is superior to other decision tree methods in model precision and recall rate and time performance is about 50% faster than the unimproved. It can evaluate tobacco comprehensive quality well.