用于调节参数区间选择的交叉验证方法  

Cross-validation Method for Tuning Parameter Interval Selection

作　　者：宁保斌 王士同 NING Baobin;WANG Shitong(School of Artificial Intelligence and Computer Science,Jiangnan University,Wuxi 214122,China)

机构地区：[1]江南大学人工智能与计算机学院,江苏无锡214122

出　　处：《小型微型计算机系统》2025年第1期104-110,共7页Journal of Chinese Computer Systems

基　　金：江苏省自然科学基金项目(BK20191331)资助。

摘　　要：现有的交叉验证方法在选择模型的调节参数时,一般在给定的参数值中选出一个最优的调节参数,而为了应对数值的复杂性和提高选择便利性,现实应用中往往更希望选出一个参数区间,也利于观察算法的稳定性.针对这个问题,本文改进了块正则化m×2交叉验证方法,提出了一个新的模型调节参数区间选择方法,基本思想是给出多个调节参数区间,采用增量的方式,不断地增加m,进而不断地减少调节参数区间个数,最终选出一个最优的调节参数区间,在这个最优区间中任意选取调节参数,都可以作为模型的调节参数.通过大量实验,与基于交叉验证的模型调节参数选择方法(m×2交叉验证方法、2折、5折、10折交叉验证)做了对比,模型在选出的区间上的平均准确度与最优单个参数的准确度相差不大,而且在该区间上最高准确度和最低准确度的差值非常小,说明在该区间上选择参数作为调节参数性能相对稳定.The existing cross-validation methods generally select an optimal tuning parameter from the given parameter value when selecting the tuning parameters of the model,and in order to cope with the complexity of the values and improve the convenience of selection,it is often preferred to select a parameter interval in practical applications,which is also conducive to observing the stability of the algorithm.To solve this problem,this paper improves the block-regularized m×2cross-validation method,and proposes a new model adjustment parameter interval selection method,the basic idea is to give multiple tuning parameter intervals,using incremental methods,constantly increasing m,and then continuously reducing the number of tuning parameter intervals.Finally,an optimal tuning parameter interval is selected,and any tuning parameters are selected in this optimal interval,which can be used as the tuning parameters of the model.Through a large number of experiments,compared with tuning parameter selection methods based on cross-validation model(block-regularized m×2cross-validation method,2-fold,5-fold,10-fold cross-validation),the average accuracy of the model in the selected interval is not much different from the accuracy of the optimal single parameter,and the difference between the highest accuracy and the lowest accuracy in the interval is very small,indicating that the performance of selecting parameters as tuning parameters in this interval is relatively stable.

关 键 词：交叉验证 模型选择 调节参数选择 区间最优 支持向量机 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]