Analytics provides trend analysis embedded in the timeseries views.
The trend is calculated using common regression methods, such as Linear regression, Exponential, Polynomial, etc.
To enable, click on the top right corner icon of the graph to open the trend analysis popup:
Trend calculation
The trend analysis is calculated based on the data point displayed. The more data points in the timeseries, the more accurate is the trend analysis and prediction.
Prediction is based on an extrapolation of the regression method. You can't predict more than 10% of the number of data points you have. Typically with 30 data points (30 days of daily averaged data), we recommend to run the prediction for no more than 2 or 3 days in the future. If your date range is the last 24 hours, you may predict for no more than a couple of hours.
Using Reporting tool, you can increase the number of data point displayed by choosing the resolution of the timeseries. For example you can display 30 days of data with a 5/15min resolution. In the dashboard view, 30 days of data will be automatically aggregated daily, hence reducing the accuracy of the trend analysis.
Resolution configuration in Reporting:
Regression methods
The best regression method depends on the shape of your data.
We typically recommend starting with Linear regression if your data follow a linear model, or Polynomial with order 2 for better fitting. The higher the order, the more the regression will fit your data. Too high, and the regression will over fit your data and reduce the accuracy at providing a general trend and prediction.
Example of polynomial regression of order 3,
Example of polynomial regression of order 6,
For documentations about the different regression methods, consult Wikipedia:
...