Stacking集成学习应用于近视矫正中的角膜塑形镜临床验配  

Application of Stacking Ensemble Learning in Clinical Fitting of Orthokeratology Lens for Myopia Correction

作　　者：巩家铭 李康妹 胡俊[2,3] 陈浩 曹倩倩 吴戈 GONG Jiaming;LI Kangmei;HU Jun;CHEN Hao;CAO Qianqian;WU Ge(College of Mechanical Engineering,Donghua University,Shanghai 201620,China;Institute of Artificial Intelligence,Donghua University,Shanghai 201620,China;Shanghai Industrial Big Data Engineering Technology Research Center,Shanghai 201600,China;Eye Hospital,Wenzhou Medical University,Wenzhou 325000,China)

机构地区：[1]东华大学机械工程学院,上海201620 [2]东华大学人工智能研究院,上海201620 [3]上海工业大数据工程技术研究中心,上海201600 [4]温州医科大学附属眼视光医院,浙江温州325000

出　　处：《Journal of Donghua University(English Edition)》2024年第2期184-194,共11页东华大学学报（英文版）

基　　金：Shanghai Science and Technology Project,China(No.20DZ2251400)。

摘　　要：针对角膜塑形(orthokeratology,OK)镜临床验配难度系数大和过程繁琐费力的问题,提出一种stacking集成学习方法预测OK镜参数值,实现OK镜智能验配。通过基于F检验的特征衍生和基于方差-改进Boruta算法的特征选择,构建出最适合目标变量的特征集合。研究了以随机森林(random forest,RF)、梯度提升决策树(gradient boosting decision tree,GBDT)和支持向量回归(support vector regression,SVR)作为第一层基学习器,线性回归(linear regression,LR)作为第二层元学习器的stacking集成学习预测模型。实验结果表明模型预测结果和临床诊断结果高度一致,验证该模型可作为一种有效的辅助临床验配方法。Aiming at the problems of a large difficulty coefficient and tedious process in the clinical fitting of the orthokeratology(OK)lens,a stacking ensemble learning model is proposed to predict the parameters of the OK lens and realize its intelligent fitting.The feature set that is most suitable for the target variables is constructed by feature derivation based on F-test and feature selection under the variance-improved Boruta algorithm.A stacking ensemble learning prediction model is studied.The model uses random forest(RF),gradient boosting decision tree(GBDT)and support vector regression(SVR)as the first layer basic learners and linear regression(LR)as the second layer meta-learner.The experimental results show that the prediction indexes of the model are highly consistent with the clinical diagnosis results,which verifies that the model can be used as an effective auxiliary diagnosis method.

关 键 词：角膜塑形(OK)镜 特征工程 stacking集成模型 参数预测 智能验配 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]