Abstract

The challenge of concept drift detection is crucial in machine learning, especially in dynamic contexts where the underlying data distribution can vary over time. For the purpose of identifying concept drift, we suggest a sliding adaptive beta distribution model (SABeDM) in this study. SABeDM combines the adaptive sliding window and beta distribution techniques to track modifications in the underlying distribution of the data stream. Several synthetic and real-world datasets are used to assess the proposed model, and it is then contrasted with cutting-edge drift detection systems. Regarding detecting true positive, false positive, false negative, and delay, our experimental results demonstrate that SABeDM works better than the currently used methods (SRP, ADWIN, DDM, and EDDM). Accuracy, precision, recall, and F1-score were also utilised as evaluation criteria. When used in a variety of applications, such as online learning, data stream mining, and real-time monitoring systems, SABeDM offers an effective and fast way to identify concept drift in a dynamic context. The proposed approach is a promising tool for machine learning practitioners to use in practical applications since it can help to enhance the dependability and accuracy of decision-making systems in dynamic situations.

中文翻译：

---

SABeDM：动态环境中概念漂移检测的滑动自适应 beta 分布模型

摘要

概念漂移检测的挑战在机器学习中至关重要，特别是在底层数据分布可能随时间变化的动态环境中。为了识别概念漂移，我们在本研究中建议采用滑动自适应 beta 分布模型（SABeDM）。 SABeDM 结合了自适应滑动窗口和 beta 分布技术来跟踪数据流底层分布的修改。使用几个合成的和真实的数据集来评估所提出的模型，然后将其与尖端的漂移检测系统进行对比。在检测真阳性、假阳性、假阴性和延迟方面，我们的实验结果表明 SABeDM 比当前使用的方法（SRP、ADWIN、DDM 和 EDDM）效果更好。准确率、精确率、召回率和 F1 分数也被用作评估标准。当用于在线学习、数据流挖掘和实时监控系统等各种应用时，SABeDM 提供了一种有效且快速的方法来识别动态环境中的概念漂移。该方法对于机器学习从业者来说是一种在实际应用中很有前途的工具，因为它有助于提高动态情况下决策系统的可靠性和准确性。