标题： 基于得分的概念漂移控制图的自助控制限 显示英文标题

标题： Bootstrapped Control Limits for Score-Based Concept Drift Control Charts

摘要： 监测由拟合的监督学习模型表示的预测关系中的变化（即概念漂移检测）是一个广泛的问题，例如，用于回顾性分析以确定预测关系在训练数据中是否稳定，用于前瞻性分析以确定何时需要更新预测模型，用于其行为可通过预测关系表征的过程的质量控制等。 一种通用且强大的基于Fisher得分的概念漂移方法最近被提出，其中概念漂移检测简化为使用多元指数加权移动平均（MEWMA）来检测模型得分向量均值的变化。 为了实现该方法，初始数据必须分成两个子集。第一个子集作为训练样本，模型在此上进行拟合，第二个子集作为外部样本测试集，从中确定MEWMA控制限（CL）。 在本文中，我们开发了一种新的引导程序来计算CL。我们的引导CL在样本量和/或误报率较小时能更准确地控制误报率。它还允许整个初始样本用于训练，从而得到更准确的拟合监督学习模型。 我们表明，标准的嵌套引导（内部循环考虑未来数据的变异性，外部循环考虑训练样本的变异性）大大低估了变异性，并开发了一种类似632的修正方法，以适当考虑这一点。 我们通过数值示例展示了优势。 显示英文摘要

摘要： Monitoring for changes in a predictive relationship represented by a fitted supervised learning model (aka concept drift detection) is a widespread problem, e.g., for retrospective analysis to determine whether the predictive relationship was stable over the training data, for prospective analysis to determine when it is time to update the predictive model, for quality control of processes whose behavior can be characterized by a predictive relationship, etc. A general and powerful Fisher score-based concept drift approach has recently been proposed, in which concept drift detection reduces to detecting changes in the mean of the model's score vector using a multivariate exponentially weighted moving average (MEWMA). To implement the approach, the initial data must be split into two subsets. The first subset serves as the training sample to which the model is fit, and the second subset serves as an out-of-sample test set from which the MEWMA control limit (CL) is determined. In this paper, we develop a novel bootstrap procedure for computing the CL. Our bootstrap CL provides much more accurate control of false-alarm rate, especially when the sample size and/or false-alarm rate is small. It also allows the entire initial sample to be used for training, resulting in a more accurate fitted supervised learning model. We show that a standard nested bootstrap (inner loop accounting for future data variability and outer loop accounting for training sample variability) substantially underestimates variability and develop a 632-like correction that appropriately accounts for this. We demonstrate the advantages with numerical examples.