Geology Reference
In-Depth Information
Fig. 2.5 Hypothetical
relation of model complexity
with sensitivity,
flexibility,
data requirements and
predictive error
Models of different complexity may show different modeling properties, such as
sensitivity,
flexibility, error, and data requirements based upon their structure.
Figure
2.5
illustrates the hypothetical relationship between model sensitivity,
modeling error, model
flexibility, training data requirement, and model complexity.
2.5.3 Training Data Requirements
More complex models may have more parameters, state variables, or linkages, and
therefore the hypothesis is that such models require more data. However, for node-
based modeling systems, if the training data length causes over
tting then the
hypothetical relation depicted above may not be true. There is a need to investigate
the optimal length of the training data phase as too little data for training may lead
to poorly trained model and too much data may lead to over
tting. The hypothetical
relation of model data requirements for node based modeling systems is as shown in
Fig.
2.6
.
Fig. 2.6 Hypothetical
relation of modeling data
length in node based models
