In order to train a robust and reliable ML model and avoid issues
regarding co-linearity , it is helpful to pick significant features for
the training process. This can be achieved by applying PCA or PLS
regression, but since the number of parameters is manageable for this
work, features are picked based on operator experience and a correlation
matrix instead (Figure 4). An advantage of this approach is that the
results will be more interpretable compared to PCA or PLS regression,
which addresses possible veracity issues as well.
The values in a correlation matrix describe the linear relationship
between two parameters, where 1 and -1 indicate a perfectly linear
relationship. The sign of the correlation coefficient describes the
direction of the relationship: a positive sign means that the value of
one parameter increases or decreases, if the other parameter increases
or decreases, respectively; a negative sign describes the opposite
relationship.