Looking Back at Forecasting in Support of Analytics

Forecasting is an important part of any kind of analytics or predictive AI.    Ultimately you have to use predictive systems not only in their current contexts but also for needed future contexts.  In other words the world,  the data it provides,  and much metadata,  is always changing.

This led me back to work we had done with forecasting methods in the enterprise, notably using the work of Wharton Prof:  J Scott Armstrong, the editor of The Principles of Forecasting   .. "

Ultimately this work led us to better forecasting methods that could be linked to some of our approaches, like the 'Business Sphere',  mentioned a number of times here.

We were particularly interested in how domain knowledge could be used (without bias and with transparency) to validate and improve such forecasts for the use as an assistant for multiple decision makers.   Including human and machine data and agents.   In real time and not.   Early on we examined rule bases to directly integrate domain knowledge.  Now we could integrate detected  'pattern knowledge' as well.    Here is a paper about using rules in forecasting.