Dan Frankowski
Aug 26, 2017 · 1 min read

Thanks for your question Jekaterina Kokatjuhha.

The mean difference between the current day and all other days would start constantly alarming if the graph grew at a constant rate. Even taking a limited past window would start constantly alarming if the graph grew fast enough. We have plenty of graphs that grow quickly. :)

There are many things we could do to improve the detection of anomalous values. They all have to do with refining our predictive model. But, it was more important to get something in place quickly and get its actual output in front of the right people to see if it would have value.

)
    Dan Frankowski

    Written by

    Data scientist, software engineer, ex-Pinterest, ex-Google, trumpet player, composer.