Time series decomposition works by splitting a time series into three components: seasonality, trends and random fluctiation. To show how this works, we will study the decompose( ) and STL( ) functions in the R language.
It can be difficult to detect the underlying trend of a time series in the presence of anomalies, due to unwanted noise. Fortunately, there are techniques to take into account those anomalies, so you can work with this kind of time series. One of them is the moving median.
Our brains are really fast at recognizing patterns and forms: we can often find the seasonality of a signal in under a second. It is also possible do this with mathematics using the Fourier transform.
First, we will explain what a Fourier transform is. Next, we will find the seasonality of a website from its Google Analytics pageview report using the R language.
Graphite is a very powerful set of tools for monitoring and graphing server performance created in 2006. Graphite is still wildly used, but it’s now falling behind more modern monitoring solutions such as the trendy solution InfluxDB / Graphana. Read More
Monitored metrics very often exhibit regular patterns. When the values are correlated with the time of the day, it’s easier to spot anomalies, but it’s harder when they do not. In some cases, it is possible to use machine learning to differentiate the usual patterns from the unusual ones.
Instinctively we look at anomalies by examining the number of events during a fixed period of time. However, this method can’t achieve fast detection rates, and fails to detect some anomalies. The Poisson distribution turns the problem upside-down by looking at the period of time for a fixed number of events .
The normal distribution is the holy grail of anomaly detection. Normally distributed metrics follow a set of probabilistic rules. Values that follow those rules are recognized as being “normal” or “usual”, while values that break them are seen as being unusual, indicating anomalies.