Anomaly detection in Power BI was introduced back in November 2020 as an update. So far, this feature, anomaly detection, is only workable for line charts, but it owns significance for report creators.

However, for the sake of new learners, it is necessary to build an understanding of anomaly detection, its applicability, and its importance. Following that, we’ll go through the 3 steps for anomaly detection in Power BI. Not only this but also, we’ll learn how to go deep down on a specific anomaly point and add it to the Power BI report.

So, here we go!

What is anomaly detection?

Anomaly detection, aka outlier detection, is a part of data mining to pinpoint unforeseen events, observations, data points, or items in your dataset that deviate considerably from the rule.

These anomalous data points can give the indication of critical events, such as a technical glitch, or prospective openings, for example, a shift in customer behaviour.

What is time series data?

To have a solid understanding of time series data, let’s break the definition into two parts. First, time series data is data which covers a chain of values or points over a period. Second, each value or point is a pair composed of two elements i.e., a timestamp for when the metric was taken, and the value related to that metric at that time.

So, when it comes to anomaly detection, efficacious detection relies on precisely analyzing time series data in real time.

What is the importance of anomaly detection?

Time series data itself never gives a whole picture. Instead, it only provides the information required for making informed guesses about what can logically be expected in the future. Whereas anomaly detection benefits by utilizing those expectations and highlighting certain indicators. Thus, it distinguishes actionable indicators within the data, discloses outliers in Key Performance Indicators (KPIs) and signals key incidents in your organization. For instance, by looking into the following time series metrics, businesses can be benefited in advance

• Web page views
• Daily active users
• Cost per lead
• Cost per click
• Revenue per click
• Bounce rate
• Churn rate
• Average order value

So, by analyzing the data associated with any of the above examples, one can take out deviations from the normal trend. Thus, these deviations, aka anomalies, in return, let’s know if this is a technical glitch or a prospective opening. In other words, the benefit is yours whether you find a glitch or an opportunity.

What are the approaches to anomaly detection?

To build a required and reliable detection system, understanding anomalies in data and the suitable choice of anomaly detection approach matters a lot. You have now a grip on the prior i.e., understanding of anomaly detection. So, to create clarity of a suitable choice of approach, here are some popular anomaly detection techniques.

1. Density-based techniques
2. Cluster analysis-based techniques
3. Bayesian Networks
4. Neural networks, autoencoders, LSTM networks
5. Support vector machines
6. Hidden Markov models
7. Fuzzy logic-based outlier detection

Since digging deep down into each approach is not the topic of this blog, we’ll soon cover them in later blogs. So, heading toward the main topic, let’s find out 3 steps for anomaly detection in Power BI.

Explore Anomaly Detection in Power BI

Let’s go through steps one by one to understand how to detect anomalies in Power BI for time-series data. We have taken a superstore’s yearly total sale dataset in this case.

Start by importing your file under the home tab, selecting your required columns for your line chart and loading it.

Step 1

Under “Visualizations”, select line chart

Step 2

Select columns for the line chart

Your line chart will appear as shown below

Step 3

Go to “analysis” under visualization

And turn on “Find anomalies”

Here is the final visualization of the line chart with anomalies

How to get a “possible explanation” of an anomaly point in Power BI?

To get a “possible explanations” of an anomaly, select any point on a line chart (with anomaly detection turned on). “Possible Explanations” will appear on the right side of the chart.

This will let you dive deep into details of a particular anomaly by digging into the category, and its sub-category i.e., it goes down each level and reveals the anomaly strength at each.

How to add details of an anomaly point to your report in Power BI?

Below the Graph of “possible explanation” of an anomaly point, the “Add to report” option is available. You can not only check details of that particular anomaly point but also can add it to your report by clicking on “Add to report”

After adding it to the report, the final report of that particular anomaly point will appear in your report as shown below

Limitations of Anomaly Detection in Power BI

• Anomaly Detection is only possible for line charts with time series data in the Axis field and with the above three data points.
• It is not workable with legends, multiple values or secondary values in line chart visual and even not with Forecast/Min/Max/Average/ Median/Percentile lines.
• Anomaly Detection is insupportable for Direct Query over SAP data source, Power BI Report Server, Live Connection to Azure Analysis Services and SQL Server Analysis Services.
• Anomaly “possible explanations” do not work with ‘Show Value As‘ options.
• Drilling down to move to the succeeding level in the hierarchy is insupportable.

TO CONCLUDE…..

Anomaly detection helps report creators highlight both positive and negative deviations in their time series data. And signals towards technical glitches and prospective openings to either mend the error or to avail the opportunity. In both ways, this small analysis can help businesses to get benefited beforehand by making informed decisions. So, develop your understanding of the concept of anomaly. Most importantly, get to know about three fast steps to find anomalies in time series data and the limitations of anomaly detection in Power BI through our blog.