Beringer Associates Technology Blog
As we move into a data-driven world, a growing number of companies are using Machine Learning software to answer their business analytic questions. Have you been looking into Opportunity sales forecasting the moment an Opportunity is opened? Have you ever wanted to provide buying recommendations for your customers as selling points? Or maybe you’ve been looking to schedule promotion calls to customers at just the right time? All of these predictive sales advantages are possible with a type of programming called Machine Learning.
If you’re following new tech trends, you’ve probably heard the term Machine Learning come up in a few places. Machine Learning is currently a popular model for predictive analysis. The “learning” portion comes from providing your Machine Learning program with a large set of data. Your program will then use that dataset to predict trends in your information. You can predict sales trends, market prices, even future medical aliments a person might have.
The data requirement for using this tool makes it difficult for smaller organizations and individuals to utilize this type of programming. Plus, constructing a Web Service that predicts trends isn’t a simple undertaking. Microsoft Azure luckily provides a platform for you to not only create Machine Learning projects in a user-friendly interface, but also shares sample data for training your projects!
In the interest of protecting company data, I’ve chosen to use a sample dataset and experiment (Sample 6) from Azure for Automobile Prices. Feel free to imagine we are demonstrating this example with your CRM Opportunity Data for Revenue Forecasting. We’ll take our Machine Learning Model from Training to Prediction Testing then to a functioning Web Service.
Get Started with a free Workspace for Azure Machine Learning here:
You will start with your experiment in a Training mode. This is where you provide the program with Data and Models for predictive analysis. You can see this experiment is comparing two types of analysis, Poisson Regression and Decision Forest Regression. You can evaluate the accuracy of each model so you know which will be best for your final web service
You can right-click the Evaluate Model Blocks for statistics on how well your model works
An R2 value close to 1.0, not too shabby! Looks like Decision Tree Regression is our pick! Once you compared and feel comfortable with your model, you can start constructing the Web Service that you can use for predictive analysis.
Select the Train Model you would like to use for the Web Service and click the “Set Up Web Service” button. I chose to create a Predictive Experiment.
Once Azure sets up your Web Service, you’re ready for testing!
The Default Configuration is usually a good representation of your Model for the Web Service, however you can tweak it if you’d like. Once you are finished, Deploy your Web Service!
Your Deployed Web Service will show up on the screen with some usage information. Azure provides your Web Service with full usage documentation, Endpoint URLs, an API Access Key for security, and sample JSON requests for testing.
You can even test your Web Service using the Test button
Want to share your Predictive Model, but don’t want to develop a REST client? Download the Excel Workbook
Azure Machine Learning Studio is a fantastic visual interface for designing a working Machine Learning program. For users who would like to experiment with a basic model for predicting data trends, this is a neat tool to look into. Features like automatic Web Service deployment and exporting the Model to Excel makes Machine Learning Studio incredibly powerful. For a basic tutorial to get you started, check out the link below. Thanks for reading!
Azure Machine Learning Experiment
Beringer Associates, a Microsoft Gold Certified Partner, is always here to provide expert knowledge in topics like these. Please contact us with any questions you may have.