Interaction Detection for Business Forecasting

• Activity Number: 173
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Dan Steinberg*+
• Companies: Salford Systems
• Keywords: predictive modeling ; TreeNet ; Salford Systems ; Machine Learning ; data mining

Abstract

Recent advances in machine learning technology make it possible to determine definitively whether interactions of any degree need to be included in a forecasting predictive model. We can thus establish conclusively, for example, for a given set of predictors, whether or not a forecasting model with interactions will outperform a model without them. Further, we can now identify precisely which interactions are supported by the data, and also the degree of interaction, even in very high dimensional data. The tools we use to achieve these results are extensions of Stanford Professor Jerome Friedman's TreeNet, developed by the authors and embedded in the Salford Systems TreeNet 2.0 Pro Ex product. We illustrate the concepts in the context of a real world regression forecasting model where we are quickly able to identify all the important interactions with a modest number of boosted tree en