Designing a Regime Dependent Artificial Intelligence System
Paradigm: Divide the cases into distinct process regimes for
which the causal relationship between predictors and predictand is expected to
differ. First forecast the regime that
will occur and then use a regime specific model to predict the variable of
interest.
Rule: Design backwards up the causality chain from the
predictand.
- What
is the predictand?
- What
processes affect the predictand?
- What
is the set of regimes that determines which of these processes will be
active?
- What
predictors can be used to forecast which regime will occur?
- What
model is best suited for predicting which regime will occur?
- What
predictors can be used to forecast the predictand in each regime?
- What model
is best suited for forecasting the predictand in each regime?
Back to Index
This page was last updated by George Young
on July 3, 2013