|
|
 |
|
DECIDE
comprises of four modules:
-
Data Pre-processing:
This is an often necessary and usually work
intensive process. You may automatically
detect and subtract a linear trend, the
mean, periodic cycles (e.g. daily, weekly,
monthly …), and you may normalise your data.
With a few treatments you transform raw data
into perfectly trimmed and meaningful
variables.
-
Pattern Recognition
is an extremely powerful method for data
compression and noise reduction. Furthermore
it eliminates colinearities in your input
data. High dimensional problems become only
feasible after a substantial dimension
reduction, using pattern recognition. DECIDE
incorporates a powerful pattern recognition
algorithm that allows you tackle highly
complex problems and to extract valuable
knowledge from your data
-
Modelling/Forecasting:
In DECIDE we implemented a non-linear,
non-parametric modelling algorithm based on
multivariate adaptive regression splines. As
of forecast precision this method yields
comparable results with artificial neural
networks but at much lower computational
costs. Multivariate adaptive regression
splines can deal with much shorter training
periods; they are far more robust, and
computationally much more efficient.
Depending on the application this method is
up to 300 times faster than neural
networks. Last but not least DECIDE
modelling results can be easily interpreted,
which provides valuable insight into the
functioning of the problem of interest.
|
|
DECIDE Toolkit