Discovering Multi-Dimensional Time Series Anomalies with K of N Anomaly Detection

In recent years there has been significant progress in univariate time series anomaly detection. However, efforts to generalize this success to the multi-dimensional case have met with limited progress. The main difficultly appears to be that in any N-dimensional time series, the anomaly will generally only manifest itself on K of the time series, with K < N. This leads to a chicken-and-egg problem. If we knew which K time series exhibited the anomaly, it would be easy to discover its location. However, we do not know this in advance, and the search space is of size 2N and not obviously amiable to greedy search. In this work we show a novel, simple algorithm that allows us to quickly find the best K of N anomaly subset for any value of K. Moreover, we show a simple metric that can rank the top anomaly subsets for all values of K from 1 to N. While our methods are mostly agnostic to the anomaly scoring model, for concreteness we use the Matrix Profile, and show that we can discover multi-dimensional anomalies that would escape detection by all current rival methods.

This website includes all the codes, data, experiments and the figures generated for the paper "Discovering Multi-Dimensional Time Series Anomalies with K of N Anomaly Detection".

Dataset

• 20_sin_data (Figure 1, 2, 3)

• toy_data (Figure 4, 5)

• MSCRED (Figure 10)

• PV_System (Figure 11)

• MGAB (Figure 12)

• Data_With_Diff_Anomaly_Type (Figure 1 in Supplementary material)

Code

Note* after clicking on the link on top of the page there is an option on how to open the file, please select 'Open with Google Colaboratory'

• TSADIS Demo

•  TSADIS algorithm (Tables 1, 2, 3 and Figures 4, 5, 6, 7, 8, 9)

• Experiment - Detecting anomalies when there are spurious time series in data (Figure 1,2,3)

• Experiment - Sensitivity of TSADIS to subsequent length (Section 6.2, 6.3, 6.4)

• Experiment - Sensitivity of TSADIS to additional dimensions (Table 1 in Supplementary Material)

• MSCRED dataset result (Table 4, Timing Result)

• PV_System dataset result(Table 5, Timing Result)

• MGAB dataset result (Table 6, Timing Result)

• Result of TSADIS on Different  anomaly type(Figure 1 in Supplementary materials) 

• Time complexity of brut force algorithm (Section 5.2)

• Experiment - Selecting the right K (Figure 14)

• Experiment - Sensitivity to noise (Figure 15)

Supplementary material

• Additional experiments

• Slides

• Additional Slides