基于深度随机森林的新型组合分类算法  

A Novel Combination Classification Algorithm based on Deep Random Forest

作　　者：任志伟 王玉德[1] 陈婷 REN Zhiwei;WANG Yude;CHEN Ting(Qufu Normal University,Qufu Shandong 273165,China)

机构地区：[1]曲阜师范大学,山东曲阜273165

出　　处：《通信技术》2021年第12期2614-2620,共7页Communications Technology

摘　　要：针对当前单分类器融合算法不灵活的缺点,提出基于深度随机森林的新型组合分类算法。该算法先建立50~400深度的随机森林,然后训练找到5个不同深度最优随机森林的模型,将其组合成一个新的模型。在威斯康辛州(诊断)乳腺癌数据集、无线定位数据集和加利福尼亚大学尔湾分校的汽车评估数据集上对提出的算法进行了实验验证,并与传统算法结果进行比较,提出的组合算法在3个数据集上分别取得了97.0%、98.0%和97.6%的准确度。实验证明,与其他传统算法相比,本文算法的准确度更高,鲁棒性更好。To address the shortcomings of the current single classifier fusion algorithm that is inflexible,a novel combinatorial classification algorithm based on a deep random forest is proposed.The algorithm first builds a random forest of 50~400 depths and then trains to find five models with different depths of optimal random forests,which are combined into a new model.The proposed algorithm is experimentally validated on the Wisconsin(Diagnosis)Breast Cancer Dataset,the Wireless Localization Data Set,and the University of California,Irvine UCI Car Evaluation Data Set,and the results are compared with traditional algorithms.The proposed algorithm achieved 97.0%,98.0%,and 97.6%accuracy respectively on the three datasets.Experiments demonstrate that the proposed combined algorithm has higher accuracy and better robustness.

关 键 词：随机森林 组合深度 集成分类 鲁棒算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]