异常检测，离群检测，波动点检测(Anomaly/outlier/

一直有这个疑问，就查询了一下，国内好像没有找到，找到了外文，翻译一下。

2022-05-22 22_04_11-What is the difference between Anomaly Detection, Change Point Detection, and Ou.png

Anomaly detection and outlier detection have the same meaning - except used in different contexts of observing data.

  - Both refer to rare events
  - Anomaly detection is often used when observing a rare event where
    -- there is no doubt about the observation itself
    -- but the event itself is not a “normal event”. Example of this is credit card fraud
  - Outlier detection is often used to refer to a rare observation where
    -- one may attribute to noise in the observation etc.
  - these differences are quite fuzzy -one could consider calling fraud detection as an outlier event too.
Change point detection is quite different from anomaly detection/outliers. It refers to changes in time series data.

异常检测和离群点检测具有相同的含义，只是在观测数据的不同上下文中使用。

• 两者均指罕见事件

• 在观察罕见事件时，通常使用异常检测

  • 这一观察结果本身是毫无疑问的

  • 但该事件本身并非“正常事件”。信用卡欺诈就是一个例子

• 离群点检测通常用于指一种罕见的观察，其中

  • 人们可能将其归因于观测中的噪音等。

• 这些差异相当模糊——我们也可以将欺诈检测称为离群点事件。

波动点检测与异常检测/异常值有很大不同。它指的是时间序列数据的变化。

搞一半天，我们搞AIOPS的异常检测，可能在时序数据上，主要指的是波动点异常检测吧。