Overview

Once you get to the pattern creation page, a regression pattern can be created using the "+ Regression" icon (1):

 

 

which will take you to the regression pattern creation page:

 

 

1. Regression pattern methods

The default regression method is the machine learning algorithm, but there are other options to choose from:

 

 

a. Machine learning

This method provides the most powerful and complex algorithm, allowing more flexibility in learning the intricate short—and long-term behaviours of your time series sensor data.

 

b. Linear

The linear option allows you to fit a linear curve over your sensor data with an upward or downward trend, following the general equation of:

 

where q is the sensor metric time series of interest, and t is time. The model fits the best values for the two parameters of the equation.

 

c. Exponential

The exponential option allows you to fit an exponential curve over your sensor data with an upward or downward trend, following the general equation of:

 

d. Harmonic

The harmonic option allows you to fit a harmonic curve over your sensor data with a downward trend only, following the general equation of:

 

e. Hyperbolic

The hyperbolic option allows you to fit a hyperbolic curve over your sensor data with a downward trend only, following the general equation of:

 

where 0 < b < 1 is the third parameter in this case.

Note that for these curve-fitting methods (b to e), the calculated parameters are available after the analysis is done in the pattern list view:

 

 

or on hover over the pattern in the graph.

 

 

2. Automatically update pattern toggle

This toggle is exclusively available when the machine learning method is selected. Once enabled, the machine learning model is automatically updated as the sensor data keeps streaming after the initial creation of the pattern. The future pattern data is inferred based on the latest machine learning model at the time of creation, making a pattern that lives and evolves side by side with your sensor data. Enabling this option will convert the end date options (see next section) to a "rolling now" so that the new data will be included as it arrives when updating the pattern.

 

3. Time range

Through this, the users can set the data time range for which the pattern should be created. When a machine learning method that is NOT automatically updated is selected, or when linear, exponential, harmonic, or hyperbolic methods are selected, the only available pattern time range is the custom date. In these settings, the user sets the fixed start and end dates where the pattern is associated, for a one-time pattern model and pattern data creation. This is more suited for performing a historical analysis on the sensor data.

 

 

When the machine learning method with automatically update pattern toggle enabled is used, the end date in all the time range selection options is rolling now, and the user may select a fixed or rolling start date using the options provided:

 

 

This is an example of a custom date with a rolling now:

 

 

If users want to ensure a high-quality pattern, especially when they know the sensor has experienced a major change in behavior or a shift in its range of values, it is recommended to set the start date to the point after the last major change. This helps the model learn from data that reflects the sensor’s current operating conditions, leading to more accurate and reliable results.

 

4. Sampling interval

Sampling interval determines how finely the data is averaged before training. 
By default, it’s set to auto, where we automatically approximate the right interval based on the sensor’s estimated logging frequency, using the data time span and number of observations. 
If you prefer, you can also set it manually by disabling the auto toggle. The options here depend on the automatically update pattern toggle. When disabled, you have more flexibility in setting the exact sampling interval for your specific data and analysis needs:

 

 

When the machine learning method is selected with the automatically update pattern toggle on, you may choose from the following sampling interval options:

 

 

The graphs below compare a 5-minute sampling interval machine learning pattern (top) to a 1-day sampling interval pattern (bottom). Notice that the 1-day sampling interval results in a model that is smoother and less affected by short-term noise:

 

 

You may further smooth the sensor and pattern data in the display panel, using the data averaging functionality:

 

This would be how a monthly data averaging looks using the original 5-minute sampling interval pattern created above:

 

 

5. Forecast

If the user is interested in predicting the possible behaviour of the sensor into the future, this configuration allows setting the duration of the forecast, starting from now:

 

 

This would draw a vertical line on the graph, marking the present moment. 
Everything to the right-hand side of this line is the forecast into the future.

 

 

6. Confidence value

This option is only available for the machine learning regression patterns. Alongside the main prediction model for the expected values, two additional machine learning models are trained to define the upper and lower confidence bands around the expected values.
 In simple terms, that band shows the range where the model expects most of the data points to fall. A higher confidence value makes the band wider, including more uncertainty, while a lower one makes it narrower and more sensitive, focusing more on the most likely outcome.

 

 

Strict setting uses a 90% confidence value, standard does 95%, and relaxed uses 99%.Here is a comparison of the strict (darker shade) and relaxed confidence values on the created machine learning regression pattern:

 

 

7. Create

Once all the desired configurations are set, click on the create button at the bottom of the page to start creating the pattern. The pattern should be ready shortly, in the list view and in the graph.

 

 

You may hide/show a pattern by clicking on the eye icon next to each pattern, edit certain fields of the pattern, or delete it using the list view.

 

Editing a pattern

Depending on the pattern method, you may have access to edit the following fields:

• Turning the automatic pattern update on or off
• Enabling or disabling the forecast
• Changing how far ahead the forecast goes

 

 

Then click on "Save" to apply the changes.

Patterns Part 1: What does the Pattern tool do?

Patterns Part 2: How to Create a Pattern?

Patterns Part 3: How to Create a Regression Pattern?

Patterns Part 4: How to Create a Trend Pattern?

Patterns Part 5: How to Select a Machine Learning Regression Pattern Method?