What algorithm can detect anomalies on the chart?

I have data on visits to a number of pages, the last 30 days. Looks something like this:

Page 1: [1,2,0,4,6,1,7,4,7]
Page 2: [3,4,12,1,7,1,2,0]


These pages very much. I need to isolate the pages that felt unusual inflow or outflow of users at any given time. What algorithm or sequence of algorithms suitable for this task best?

UPD: while looking at the machine learning algorithm anomalies detection, but maybe there's a way to quickly, for example (thinking out loud) can break the data sets into several equal parts and compare them to interest fluctuations. If all in the range 0, then we can assume that there are no abnormalities, the leap - then something went wrong. Most likely, it will do.
June 3rd 19 at 19:20
4 answers
June 3rd 19 at 19:22
Solution
It is possible to calculate the variance in a certain period of time (window) and if the value falls outside the average +/- 3σ, perhaps, is the anomaly.
https://www.slideshare.net/YoshihiroIwanaga/anomal...
https://stackoverflow.com/questions/2303510/recomm...
June 3rd 19 at 19:24
This approach will give nothing.
Study cohort analysis and Lean metrics. They are called AAARR.
Training on the anomalies will work fine. - shea.Be commented on June 3rd 19 at 19:27
June 3rd 19 at 19:26
Anomaly detection in network monitoring... - theory, the name of the algorithm, useful links
June 3rd 19 at 19:28
You can, of course, to invent the Bicycle. But you can get mind-mind, starting with theory. The more you in good stead, because the problem that you describe is found in different types in the economy, information security, medicine, technical diagnostics, marketing - including anomalies of visits to pages, type of your - and more in dozens of other subject areas, and studied this task you will provide a real interest in you as an expert dozens of employers in the future.
This theory is called really different - "search and anomaly detection", "changepoint detection", "detecting discords and emissions", etc. In the first approximation it all comes down to time series analysis and methods of classification, and change detection models, which describe the data ( "exceeding thresholds", 3сигма, etc. - is only the most trivial and naive of the methods that are used today. But of course not the "interest rate fluctuations"). Moreover, if you want to make it serious, it is necessary to study the parameters of the alignments themselves (not only average and variance), to test korrelirovannykh visit the pages of sites, to detect trends and seasonality, to check for clustering in the data etc ..
Well, you can, of course, "quickly", just have something there quasione considered. Though shown to a customer. Then Yes - have calculated, our deviations, I drew a beautiful chart, impressed the customer, he received remuneration, profit. Everyone chooses their own path.
no, fine here you. You asked "Which algorithm can detect anomalies on the chart," you answered. But really, you wanted "Make for me". That's why you and stackoverflow is like, not because they gave you the answer, but because they did for you that you just emphasize. And then there are people who after such answers create questions in a week or two and ask why it isn't working as hoped, and how to configure it. - shea.Be commented on June 3rd 19 at 19:31
Nurlan@daager, Alas, here the author of the question - The Whiz - quietly deleted my comment that you very accurately answered. Therefore, your comment as if "hung in the air." I don't think such a move is The Whiz ethical correct, but everyone has their own idea about it. - Jeff commented on June 3rd 19 at 19:34

Find more questions by tags AnalystMathematical statistics