Why checking your KPI several times a day? To detect problems as fast as possible. Of course, you can’t spend your time simply watching the KPI all day long.
Running a business successfully you have to monitor some important Key Performance Indicators. For example, you should always look at your lead generation KPI several times a day.
Monitoring KPI, sales or any product you’re looking at an ecosystem. In an ecosystem, things behave in harmony (kind of). Very often you will see your metrics correlating with each other. When a usual correlation between two metrics gets broken we can suspect something strange is happening.
Let’s look at anomaly.io analytics at its beginnings ( a long time ago ). See the above graphic. In green the new users and in red the returning users. Clearly, something strange happens in the middle of November. But let’s use some techniques to detect more!
To detect the correlation of time series we often use the auto-correlation, the cross-correlation or the normalized cross-correlation. Let’s study some techniques and understand them better.
Previously, I tested the Twitter Breakout (EDM) to detect Anomalies. As the popular E-Divisive, EDM detects mean shift and change in distribution. Both algorithms work ok with seasonal time series but, perform even better without the seasonality. Read More
Time series decomposition split a time series into a seasonal, a trend and a random residual time series. The trend and the random time series can both be used to detect anomalies. But detecting anomalies in an already anomalous time series isn’t easy.
What is a seasonal adjustment? Why and when to use it? How can it be calculated? A seasonally adjusted time series is a time series with a removed seasonality. To compute this metric, we need to calculate the seasonality. Then, we remove it from the original time series.
How is the time series decomposition working? How does it split a time series into three other: Seasonality, Trends and Random? To understand time series decomposition, we will study the decompose( ) and STL( ) function from R language.
It might be difficult to detect the underlying trend a time series follow in the presence of anomalies. It often has unwanted noise. Hopefully, there is some technique to overpass those anomalies in order to work with this kind time series. The Moving Median is one of them.