Late last year Microsoft expanded their Power BI capability to use automated machine learning (ML) directly in Power BI. This means that business analysts can now train, validate and invoke ML models within the Power BI app without having the technical knowledge of data scientists.
It offers three models:
- Binary Prediction, which classifies the outcome into two categories, for example, True or False, Positive or Negative;
- Classification, which categorises and assigns an observation to a group, for example classifying customers into sub-categories, and
- Regression Models that predict a numeric result based on the relationship between dependent and independent variables, for example, predicting the number of customers that will visit your site based on past numbers and their behaviour in time.
The setup will guide you to select the data you want to use. Firstly, you will have to add a machine learning model to your Power Bi, then choose what historical data to use and the outcome field for the prediction.
You will also be required to choose which model from the ones above you want to apply and data fields that it should use. Then once you have given the model a name and saved its configuration, it’s time to train the model.
Training involves up to 50 iterations with different algorithms and setting parameters. Afterwards it will tune and validate the model in the retraining phase. Once you are satisfied with the results, you can apply and publish the model within your PowerBI dashboard. AutoML will automatically generate a report that explains the performance of the model you just used. You will also be able to see statistical summary and training details in other pages of the report.
ML models are currently available to Premium and Embedded hosted dataflows users.
To sum up, some of the advantages of Power BI’s Machine Learning Integration are:
- Using ML models without being an experienced data scientist,
- No coding,
- It is simple to use and the setup wizard explains what each model does,
- Pre-built reports to easily illustrate model performance.
All of the disadvantages are related to the ML models being run by a person who is not a data scientist:
- Data often requires pre-processing to fit the model, so it will still need to be done by an experienced person,
- Not every model fits every data, therefore, having a data scientist is still important,
- To someone who is an expert, high performance percentage can mean that the model is performing well, however, just as before, a data scientist is needed to ensure that the prediction is accurate.
So, there’s our summary of Power BI’s ML integration. Hopefully you can see the potential these tools have and are keen to start using them! If you have any questions, please feel free to reach out to us at firstname.lastname@example.org for more information.
Mar 10, 2020
Author: Katrina Biseniece, Business Analyst, Blackbook.ai