Machine learning algorithms are widely used for data analysis nowadays. These algorithms collect the features and predict the trend in data. Microsoft Power BI provides several ways to use machine learning methods for predictive data analysis. To apply ML methods into the Power BI reports, we can build the ML models in Azure machine learning studio, Power BI workspace or Power BI Python script.


Azure machine learning studio

When using Azure machine learning studio for model building, we can choose to write our own code in Jupyter notebook, or use the prepared assets in designer and autoML. No matter which way we choose, we need a compute instance first by clicking Compute at the sidebar menu. The difference between running the code in Azure notebook and in the local machine is that, using Jupyter notebook in Azure requires the information about which subscription, compute, workspace and experiment to use. This information should be written in Jupyter Notebook cells. The result of the training process will be shown at experience — runs.

In the designer, different components respond for data preparing, data training, model evaluation, and score. We can simply drag and drop these components to canvas step by step and fill some conditions in settings, then deploy the model. Using automated ML is simpler, since we don’t need to decide which ML algorithm to use. We can select the dataset, create or generate the experiment then choose the task type of the training. The system will train the data using different algorithms. We can find the training results at the model tab and deploy the best model for our dataset. After the model is trained and deployed, it can be consumed for power bi. Select transform data from power bi and then AI machine learning at AI Insights. Then we can find all the models we deployed in Azure machine learning studio. Select a model, then we can get the predicted value by this model.


Power Bi dataflow

Power BI workspace also provides auto ML methods. By selecting dataflow under new at power bi workspace, we can create new entities and import training and test dataset. Next step is to select Machine learning model and choose the training data, predicted column, the related training columns, and training time. The training algorithms can be changed at another training model. The longer you train the model, normally the better prediction result it has. After training is finished, the predicted result can import to Power BI reports by get data—power bi dataflows.

Python script

Python script is another option to build a machine learning model in Power BI. The machine learning algorithm library such as scikit-learn and data process library pandas are supported by power bi. Select python script in transform data, then we can implement the code for generating the prediction data.

All in all, we can write our own code or use auto ML model to apply machine learning algorithms in Power BI. Azure machine learning studio creates the weight of the model we choose, and Power BI workspace generates the predicted data directly. Writing our own code provides more flexibility, while automated ML methods are easy to use for those who don’t want to code.


Need support implementing machine learning functionality in your Power Bi environment?

At HUBSTER.S we can help you achieve your Organizational Data Strategy by building the solution you need to connect different areas of your business and explore endless possibilities of an integrated self-service BI system. Our employees are experts in Business Intelligence, Data Analytics, Data Warehousing, Machine Learning, Production Management, Finance and Controlling. If you need support in the implementation of your BI or Analytics project, please contact us at any time!

Linus Trips HUBSTER.S

Qianyu Chen

Qianyu Chen ist Solution Architect für Data Analytics und Machine Learning.