Identify Data Anomalies

Identify Data Anomalies

Anomaly detection improves line charts by automatically recognizing abnormalities in time series data. It also explains the abnormalities to assist with root cause analysis. Without chopping and parsing the data, you can simply get insights with a few clicks. Anomalies may be created and seen in both Power BI Desktop and the Power BI service. This article’s methods and graphics are from Power BI Desktop.

Get started

1. This lesson makes use of internet sales data for a variety of items. Download the example file of an online-sales scenario to follow along with this course.
2. Anomaly detection may be enabled by choosing the chart and then finding anomalies in the analytics pane.
3. For instance, it depicts revenue over time. When you include anomaly detection, the chart is immediately enriched with anomalies and the anticipated range of values. When a value deviates from the predicted range, it is labeled as an anomaly. 

Anomalies in format

• This experience may be completely customized. You may change the shape, size, and color of the anomaly, as well as the color, style, and transparency of the predicted range. You can also change the algorithm’s parameters. The algorithm becomes increasingly sensitive to changes in your data as the sensitivity is increased. In such a situation, even little deviations are recorded as anomalies. When the sensitivity is reduced, the algorithm becomes more discriminating in what it deems an abnormality.

Explanation

• You can not only discover but also automatically explain abnormalities in data. When you choose an anomaly, Power BI does an investigation across fields in your data model to determine potential reasons. It provides a plain language explanation of the anomaly as well as aspects related to that anomaly, organized by explanatory strength.

Create explanations

• You can also choose which fields are utilized for analysis. By dragging Seller and City into the Explain by Field Well, for example, Power BI limits the analysis to to those fields. In this situation, the anomaly on August 31 appears to be linked to a certain vendor and cities. Here, vendor “Fabrikam” has a 99 percent success rate. Power BI measures strength as the ratio of the departure from anticipated value to the deviation in total value when filtered by dimension. For example, for the anomaly point, it is the ratio of actual minus predicted value between the component-time series Fabrikam and the aggregate time-series total Revenue.

Considerations and constraints

1. Only line chart visualizations using time series data in the axis field facilitate anomaly identification.
2. Anomaly detection is not supported in line chart visuals with legends, multiple values, or secondary values.
3. A minimum of four data points is required for anomaly detection.
4. Anomaly detection does not operate with Forecast/Min/Max/Average/Median/Percentile lines.
5. Direct queries against SAP data sources, Power BI Report Server, live connections to Azure Analysis Services, and SQL Server Analysis Services are not supported.
6. The ‘Show Value As’ choices do not function with Anomaly Explanations.
7. Drilling down to the next level of the hierarchy is not permitted.