Once forecasting is enabled on a check, the forecaster will look at the data available from that check. The forecaster requires some data history to be available in order to even consider if a forecast is possible. Please make sure that you have at least a few days worth of data available - otherwise the forecaster will simply show you a "no forecast available" message in the web interface.
Once sufficient data material is available, the forecaster will examine these data looking for various properties, such as "periodicity", to assess whether the data is chaotic, or more or less repetitive in its nature.
The forecaster will then attempt to "model" the data set. This means, to generate a mathematical model that can re-produce the known past observations. This is a rather complicated process, and it can take many minutes (even hours on slow computers) for it to construct this model. If no suitable model can be found, the forecaster will show that "no forecast is available" in the web interface. In this case it will re-try the modelling every few hours, as new observations become available to it.
Once a model is successfully built, the forecaster will produce the actual forecast based on that model. The generation of the forecast is a relatively simple operation, once the forecast model is available. The forecast will be updated every 30 minutes, to reflect the latest changes in the observed data.