Overview
You can make data predictions directly from Explorer. This feature allows you to compare predicted values with actual values to detect potential anomalies in your data or predict when values will cross a specific threshold.
Making a Data Prediction
Warning : The data prediction feature is effective only for curves that show cyclical behavior over a fixed period. For non-cyclical curves, its use will not be appropriate and may produce unreliable results.
Connect to Io-base and access the Indaba Explorer menu.
Your prediction will be based on the data for the period selected for the displayed curve.
Make sure to select the period you want to use as a reference.
Once your curve is displayed, click on the gear icon located at the top right of your screen.
Start by duplicating the selected metric, so you can compare the curve of actual data with that of predicted data.
Next, select the “Holt-Winters” aggregation for the duplicated metric.
Several parameters will appear to configure your prediction:
To fill them out, observe the curve on which your prediction will be based :
First parameter:
Observe the curve on which the predictions are based and identify a trend. In our example, we observe the following trend :
Next, identify the significant points of this trend. In our example, the following points are noted :
Observe the times at which these points occur and calculate the difference (in duration) between these two times.
In our example, the first significant point of the trend occurs at 09:14:04 and the last significant point of the trend occurs at 09:19:49, which results in a difference of 00:05:45 (5 minutes and 45 seconds).
Therefore, enter "5 minutes and 45 seconds" for this first parameter.
"Pattern" parameter:
We observe that in each iteration of the trend, there are two points of variation.
The pattern parameter will thus be set to 2.
"Points" parameter : This parameter specifies the number of points you want to generate in the prediction.
Note: The maximum number of predicted points is limited to 1000.
"Offset" parameter :
As previously mentioned, the prediction model will be based on the data displayed in the curve.
To ensure the model is accurately predicted, position yourself at the beginning of a trend cycle.
When the value at the beginning of the curve does not correspond to the start of the cycle, you can use the offset parameter to adjust the selected period on the graph to the start of the cycle.
For example, imagine the displayed curve looks like this :
We see that the beginning of the curve does not match the start of a cycle :
We would like the curve to start at the point marked below (start of the cycle) :
Simply look at the date when the cycle start point occurs and calculate the difference with the current start date of the curve :
09:19:49 - 09:16:54 = 00:03:05
You would then enter "00:03:05" for the offset parameter to ensure the curve starts at the beginning of a cycle.
After completing this configuration, click on “Apply all settings.” Your prediction curve (highlighted below) will be displayed :
If you want to overlay the prediction curve on your actual data curve, return to the configuration screen and check the "Full Display" box :
For more details, refer to the Influx documentation :