基于模糊决策的随机森林算法

Algorithm of random forest based on fuzzy decision

为减轻现有随机森林算法中重要度低的属性对分类结果的影响,提出一种基于模糊决策的随机森林算法(FRF)。根据随机森林中对决策树设置的属性重要度的值,舍去重要度较低的属性,把完全生长的决策树转化为模糊决策树。由叶子结点中各个类别被赋予的相应权重和类别分布情况,计算得到能够判定样本类别的概率;进行随机森林算法优化,提高分类正确率。实验结果表明了优化后的随机森林算法方法在UCI标准数据集上的有效性,分类正确率得到了提高。

A random forest algorithm based on fuzzy decision was proposed to solve the influence of low-importance attributes on classification results in existing random forest algorithms.According to the setting value of attribute’s importance in the decision tree of the random forest,the fully grown decision tree was transformed into a fuzzy decision tree by discarding the attri-butes with lower importance.The probability of determining the samples class was calculated according to granted weight and distribution of each class in the leaf nodes.The random forest algorithm was optimized to increase the classification accuracy.Experimental results show that the optimized random forest algorithm is effective on the UCI standard data sets and can improve the classification accuracy.