In this paper, some significant efforts on fuzzy oblique decision tree (FODT) have been done to improve classification accuracy and decrease tree size. Firstly, to eliminate data redundancy and improve classification efficiency, a forward greedy fast feature selection algorithm based on neighborhood rough set (NRS_FS_FAST) is introduced. Then, a new fuzzy rule generation algorithm (FRGA) is proposed to generate fuzzy rules. These fuzzy rules are used to construct leaf nodes for each class in each layer of the FODT. Different from the traditional axis-parallel decision trees and oblique decision trees, the FODT takes dynamic mining fuzzy rules as decision functions. Moreover, the parameter δ, which can control the size of the tree, is optimized by genetic algorithm. Finally, a series of comparative experiments are carried out with five traditional decision trees (C4.5, Best First Tree (BFT), amulti-class alternating decision tree (LAD), Simple Cart (SC), Naive Bayes Tree (NBT)), and recently proposed decision trees (FRDT, HHCART, and FMMDT-HB) on UCI machine learning datasets. The experimental results demonstrate that the FODT exhibits better performance on classification accuracy and tree size than the chosen benchmarks.

在本文中，已经对模糊倾斜决策树（FODT）进行了一些重要的努力，以提高分类的准确性并减小树的大小。首先，为消除数据冗余并提高分类效率，提出了一种基于邻域粗糙集（NRS_FS_FAST）的前向贪婪快速特征选择算法。然后，提出了一种新的模糊规则生成算法（FRGA）来生成模糊规则。这些模糊规则用于为FODT每一层中的每个类构造叶节点。FODT与传统的轴平行决策树和倾斜决策树不同，它采用动态挖掘模糊规则作为决策函数。此外，参数δ通过遗传算法对树的大小进行控制。最后，使用五种传统决策树（C4.5，最佳第一树（BFT），多类交替决策树（LAD），简单购物车（SC），朴素贝叶斯树（NBT））进行了一系列比较实验。 ，以及最近在UCI机器学习数据集上提出的决策树（FRDT，HHCART和FMMDT-HB）。实验结果表明，与所选基准相比，FODT在分类精度和树大小上表现出更好的性能。