How to access the feature

This feature can be used for both the Graph Module and the Graph Card in Dashboards.

You can access it in the configuration for each respective graph and its sensors. 

Overview

Patterns help you model, understand, and even forecast a sensor's behavior over time. They come in two main types:

• 
Regression, where the goal is to build a curve or machine learning model that best fits the historical data, is constrained by the complexity of the model
• Trend, which looks beyond short-term fluctuations to highlight the long-term behaviour of the metric.

Regression patterns are great for forecasting or detecting deviations, while trend patterns help you see the underlying direction over time.

 

How does the pattern pre-processing work?

This section will provide behind-the-scenes information about how the tool works so that users can better understand the tool and trust their analyses.

 

Sampling Interval

The sampling interval is the amount of time between two consecutive data points in a time series used for the analysis. It averages each sensor's data over each period independently.​The example below shows a sampling interval set to 1 hour.

 

 

 

 

 

Outlier removal

Once resampled, we exclude outliers using a Hampel filter so that a bad data point doesn’t distort the model. This creates a sliding window of size 10 that flags points whose deviation from the local median exceeds 6 times the scaled Median Absolute Deviation (MAD):​

MAD=median(∣xi −median(x)∣)

 

 

 

What are the supported regression pattern types?

Regression patterns are available in two main categories: machine learning and curve-fitting.

The machine learning method

The machine learning regression algorithm learns from the cleaned historical values of a sensor metric, together with time-related features like the hour of the day or the day of the year. This helps the model capture both seasonal and overall behaviour of the data. The model is first trained on historical data to learn how your sensor behaves. It’s tested on unseen data to check accuracy before being finalized using all available history. This process ensures the model fits your data as accurately as possible.

Instead of giving only one forecast, the model can also estimate a range of possible outcomes, for example, showing values between the 5th and 95th percentiles. This helps indicate how confident the model is in its predictions.

Here is an example of time-series data:

that might get a machine learning regression pattern, showing the expected, lower, and upper band confidence intervals:

 

Curve-fitting options

The supported curve-fitting options are more traditional mathematical models, commonly used to fit a curve to declining time series data. We support several curve types: Linear, Exponential, Harmonic, and Hyperbolic. Linear and exponential curves can capture both upward and downward behaviours, while harmonic and hyperbolic are designed for declining trends only. Here is an example of exponential, harmonic, and hyperbolic curves fitted over a declining time series data:

What are the supported trend pattern types?

Trend patterns are available in two main categories: machine learning and the legacy linear trend.

The machine learning method

The machine learning trend algorithm creates a seasonality analysis that decomposes time series data into long-term trend and seasonal components, like daily, weekly, or yearly cycles. Then we only present the trend component after this decomposition. Unlike a simple linear trend-line, this model can handle non-linear behaviours and recurring patterns, making it ideal for uncovering how a metric truly evolves. This is an example of the machine learning trend created for a declining time-series sensor data:

Linear trend

This is the legacy linear trend option that fits a line to the data, for a simple and fast trend analysis of the time series sensor data.

 

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?