Earning him the coveted title of best forum speaker, Pieter van Staden, senior quantitative analyst for Nedbank Retail, recently presented his views on predictive modelling with SAS Enterprise Miner at the SAS User Forum in Johannesburg.
Delving into how to do effective predictive modelling when you are not an experienced mathematician, van Staden focused on demonstrating the simple steps involved in solving a problem typical to retail banking.
“Predictive modelling has a reputation of being the playground for experienced mathematicians and elite programmers but I really believe SAS Enterprise Miner has come to alter that perception,” says van Staden.
“The bottom line is that when you do predictive modelling you will inevitably be wrong a lot of the time, but occasionally you will be right too, and that is what counts.”
Van Staden highlighted the fact that with predictive modelling you are essentially predicting the probability of an outcome. To do this in the simplest way, there are six steps to follow.
* Step one is to establish what the actual business problem is and whether or not predictive modelling is the best tool to solve this problem.
* Step two is to assemble all your data, know what data you need and assemble it in a manageable and effective way.
* Step three is to decide on a model design and create the model. Being the most in-depth step, this is where Enterprise Miner really proves itself through the easy, visual, drag-and-drop interface to model design. Although countless possibilities exist, the example described used only eight Enterprise Miner nodes, including the non-standard credit scoring nodes, like interactive grouping and scorecard. The use of these nodes outside of a typical credit risk environment was demonstrated.
* Step four involves the actual model implementation and, inevitably, the moment of truth for your predictive modelling implementation.
* Step five focuses on the automation and maintenance of the model implemented.
* Step six is focused on adding value to the model though maintenance.
“Predictive modelling can really be made simple by following logical steps and with the help of SAS Enterprise Miner. There really is a simple, practical way to obtain useful results from predictive modelling, and the important thing to remember is that it is not an exact science,” ends van Staden.
Share