I just started exploring Anomaly detection in timeseries for Univariate, Multivariate timeseries.
I read few articles about it, few research papers as well. But every article/research paper has solution to their specific problem which i understood.
And now I really got confused what anomalies are, are anomalies just an outlier? which method to apply for Anomaly detection in general.
Can someone guide me on this?
I want to explore statistical method only for anomaly detection.
Can someone please list down statistical algorithms related to both univariate & multivariate anomaly detection timeseries.
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Generally speaking, anomalies are datapoints that deviate heavily from what is expected. There's a lot more to it than just that though.
As this has already been answered, I direct you here. As the link mentions, anomalies can be thought of in lots of different ways and which way you choose to view them is dependent on your problem. They depend on the amount of data you've captured, the mechanism that generates that data, what you plan to do with the anomalies that you do detect, etc.
If you are just looking to get started with some simple anomaly detection methods for time series I encourage you to first use your intuition about the problem you are dealing with. For example, if you are dealing with data that is extracted from a physical sensor you could have multiple types of anomalies you want to think about. Perhaps there are physical constraints to the thing that your sensor is measuring and values that end up outside of these constraints are themselves anomalies because they are unexpected by our model of the physical world. Perhaps there are jumps between consecutive values that would indicate something "strange" going on that would warrant intervention or observation of some sort. It's tough to say without understanding the problem first.
