Prime-Time: Symbolic Regression Takes Its Place in the Real World 247
Fig. 4 The founders effect means that each model development follows its own path through
search space. Here we look at the variables which have emerged to the fore. Because of the
couplings of inputs, we can see that some inputs are clearly desirable. Others can clearly be ignored
and some may be substituted for others and, in some circumstances or combinations, become useful
Fig. 5 This is a table of diverse 3-variable ensemble models for a distillation column
based on explanatory variables (economic drivers), the most popular of which
is the Auto-Regressive Integrated Moving-Average with eXplanatory variables
(ARIMAX) model. A limitation of this approach, however, is the assumption of
linear relationships between the explanatory variables and the target variable.
One option to overcome this limitation is by integrating GP and ARIMAX
methods into a hybrid system where the nonlinearity is represented with transforms,
generated by GP and used as explanatory variables in ARIMAX models.
The final forecast is generated by ARIMAX models with nonlinear explanatory
variables (Kordon 2014 ). The nonlinear equations, generated by GP, can be based