基于改进随机森林优化算法在医疗数据中的应用研究  

Research on application of improved random forest optimization algorithm in medical data

作　　者：朱城 苏前敏[1] 郭晶磊 沈宙锋 ZHU Cheng;SU Qianmin;GUO Jinglei;SHEN Zhoufeng(School of Electronic and Electrical Engineering,Shanghai University of Engineering Science,Shanghai 201620,China;Shanghai University of Traditional Chinese Medicine,Shanghai 201203,China)

机构地区：[1]上海工程技术大学电子与电气工程学院,上海201620 [2]上海中医药大学,上海201203

出　　处：《智能计算机与应用》2021年第8期130-134,共5页Intelligent Computer and Applications

基　　金：上海市2017年度“科技创新行动计划”基金资助项目(17401970900)。

摘　　要：本文针对临床疾病预测过程中医疗临床数据特征因子相关性较强,模型调参工作较为复杂的问题,提出一种多模型复合优化方法。考虑到过多的强相关性特征因子很容易降低模型的运行效率,利用SelectKbest变量筛选算法对临床数据进行筛选,采用遗传算法对随机森林分类器模型的参数选择进行优化。最后,以乳腺癌的临床数据为例,实验证明通过以上方法优化后算法模型的精准值、召回率、F1分值、AUC值等方面均有提高,该提出的超参数调优方法为具有强共线性的临床数据处理和疾病预测提供了一种新思路。In the process of clinical disease prediction,the medical clinical data feature factors are strongly correlated and the model adjustment work is more complicated.A multi-model compound optimization method is proposed.Considering that too many strong correlation feature factors can easily reduce the operating efficiency of the model,the SelectKbest algorithm is used to screen the clinical data,and the genetic algorithm is used to optimize the parameter selection of the random forest classifier model.Finally,taking the clinical data of breast cancer as an example,experiments prove that the accuracy of the algorithm model after optimization by the above method,recall rate,F1 score,AUC value and other aspects are improved.The proposed method provides a new idea for clinical data processing and disease prediction with strong collinearity.

关 键 词：遗传算法 kbest变量筛选 随机森林 超参数调优 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]