What method is used to detect an anomaly in a random sequence?

Question

data mining

Hello. There is a random (non-stationary?) time series with two variables
271-547
25-741
644-290
188-638
311-56
961-212
"714-204"
"728-209"
"714-209"
466-"204"
275 -668
528-735
948-466
203-455
848-288
317-909
449-736
Sometimes outliers (anomalies?) occur (marked in quotation marks) in the form of a sequence of not very different numbers. The length of such sequences is varied. Advise what methods can be applied to the discovery of these sequences. Thank you.

Answer 1

Eh, I saw the question too late - you gave it the wrong tag.
Here they have already thought up and advised this .... While the task you have is absolutely classic, well studied, described and even included in textbooks. Another thing is that there are many methods for solving it - depending on the characteristics of the data you are working with.
What you want to do is called "search for anomalies in time series". This phrase is easy to google. To enter the topic, you can start, for example, from here:
https://dyakonov.org/2017/04/19/search-anomalies-ano...
or from here
https://www.datascience.com/learn-data -science/fun...
There are also more serious descriptions. If you're interested, I'll let you know.
PS I forgot to say - the correct tags for your question are "Machine learning", "Data science", "Mathematical statistics", "Data mining", well, maybe with a big stretch - "Neural networks".

Answer 2

Without any additional information, you have only one method - to manually select the data, thus creating a training set for the neural network, and train it based on them.
In some cases, it’s more efficient to work not with the data itself, but with some calculated parameter based on them, well, in your case, the difference between adjacent values ​​​​or think of something else, this can make it easier to train the neural network, and in some cases it won’t even be needed, and you can formulate the condition yourself already on their basis.