Gradient Boosting Machine Analysis Template for Spotfire® 1.0

Overview

This template is used to create a GBM machine learning model to understand the effects of predictor variables on a single response.  Examples of business problems that can be addressed include understanding causes of financial fraud, product quality problems, equipment failures, customer behavior, fuel efficiency, missing luggage and many others.

You can import your own data into this template, configure and perform the analysis, evaluate the model quality and visualize the results.  It is designed for use by a business analyst or citizen data scientist.  No specialized knowlege of statistics, data science or programming is assumed.

The GBM implementation in this template has the following features:

• Accepts one numeric or binary (O/1) response (output) variable.  Some common binary responses are good/bad or pass / fail classifications.
• Accepts numeric or string predictor (input) variables
• Visualize nonlinear response-predictor relationships
• Visualize predictor interactions
• Model can include a large number - hundreds or thousands - of predictor variables.
• Results rank the most important predictor variables needed to accurately model the response.  Less relevant or redundant predictors have lower variable importance ranks and can often be ignored.
• High prediction accuracy
• Handles missing data
• Usually not necessary to filter outliers or transform data

* The Gradient Boosting Machines (GBM) machine learning model used is an ensemble decision tree method.  It is implemented using the CRAN gbm package within the Spotfire interface.  For more information on machine learning algorithms and use cases, visit the Spotfire Community Machine Learning community article. 

          Configuring and Evaluating the Model

           GBM Results:  Individual Predictor Effects on the left and Predictor Interaction Effects on the right