I am compelled by recent great work by our team at Othot to write about the potential significant impact of feature engineering and granularity. Recently, we have been exploring and understanding the impact of breaking down the features we use in our predictive models into more granular features. For example, many companies and organizations record information about visits to their sites – think about a car dealership recording that a prospect visited their show room or a potential customer visited their website. Our models show that behaviors like these are often strong indicators of a purchasing decision – especially when we consider frequency and length of visits.
Such an insight is quite expected and somewhat ho-hum, obviously. However, recent work completed at Othot shows that when we classify the visit and distinguish one visit from another we often observe additional, dramatic impact on the accuracy and stability of a model. This breakdown is not obvious and requires art behind the science. To me, this highlights the difference between “data” and “intelligent data”. Feature engineering – being thoughtful about the data – leads us to this intelligent data. Another example of obvious data is distance. If distance from point A to point B is important, then distance is the data. Intelligent data is understanding travel time, weather, and transportation options. All data is not created equally.
This is exciting work that often delivers exciting insights – resulting in refined probability scores related to whether or not a prospect will make a desired purchase decision. More importantly, these insights allow us to deliver powerful “what if scenarios” to our customers that allow them to target specific prospects with meaningful suggested actions.
Contact us at sales@Othot.com to learn more about how Featured Engineering can lead to Intelligent Data and Powerful What-If Analyses.
By Andy Hannah | March 26, 2016