基于GAMLSS-Akaike-Weights-Scoring的异常时间序列检测
Detecting Anomalous Time Series by GAMLSS-Akaike-Weights-Scoring
摘要:
提出了一种基于惩罚似然分布回归的异常时间序列检测方法。具体地,用于位置、尺度和形状的广义可加模型用于推断由具有由基函数组成的参数的参数分布定义的样本路径表示。赤池权重,然后应用到每个模型和时间序列,产生的概率测度,可以有效地用于分类和排名异常时间序列。一个数学阐述也给出了合理的建议赤池权重评分下一个合适的模型嵌入作为一种方式来渐近识别异常时间序列。在多个模拟和真实世界数据集上评估该方法的研究也证实,可以获得高精度检测许多不同和复杂类型的形状异常。这两个代码实现GAWS运行在本地机器上和本文中引用的数据集都可以在线获得。
英文摘要:
An extensible statistical framework for detecting anomalous time series including those with heavy-tailed distributions and non-stationarity in higher-order moments is introduced based on penalized likelihood distributional regression. Specifically, generalized additive models for location, scale, and shape are used to infer sample path representations defined by a parametric distribution with parameters comprised of basis functions. Akaike weights are then applied to each model and time series, yielding a probability measure that can be effectively used to classify and rank anomalous time series. A mathematical exposition is also given to justify the proposed Akaike weight scoring under a suitable model embedding as a way to asymptotically identify anomalous time series. Studies evaluating the methodology on both multiple simulations and real-world datasets also confirm that high accuracy can be obtained detecting many different and complex types of shape anomalies. Both code implementing GAWS for running on a local machine and the datasets referenced in this paper are available online.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。