Alex presented a well structured case for why decision applications should include knowledge extracted from data in addition to knowledge captured from experts. Alex used a fish processing plant as the illustration for his presentation – an interesting choice.
If you have enough data relative to your problem available to you for processing, applying predictive analytics modeling enables you to extract knowledge in the form of risk scores, fitness scores, etc… that can then be leveraged by business rules. To do so, you use standard predictive model building approaches, using tools such as the free ones provided by Project R, or one of the commercial ones.
However, one of the key issues that you will face is how to deploy the models created through the predictive analytics model building. Alex referred to typical model update processes in financial services, where the frictions in the deployment process result in 6 months+ release cycles. I am very familiar with those cases – as a disclosure, Alex and I both worked at HNC a few years ago.
Alex has been very active in the efforts to solve this deployment problem, and has contributed significantly to the creation and growth of the PMML standard. PMML is an interesting effort – it has resulted in its first generations in a well supported model description tool, but suffered from the lack of support for expressing variables. That was a key barrier to adoption, and one of the key difficulties we had to work on when designing and implementing the PMML support for Blaze Advisor.
That problem has been largely (and possibly completely – I have not checked the details) in the latest version of the standard – which is not yet supported by everybody but will be.
Alex wrote a book on PMML: PMML in action. He does cover PMML 4.
Zementis, the company Alex works for, combines Drools and a PMML deployment capability, to let you create decision applications that combine rules capturing the expertise and models managing the risk and are deployed to the cloud.
Learn more about Decision Management and Sparkling Logic’s SMARTS™ Data-Powered Decision Manager