W. K. Hardle, R. A. Moro, and D. Schafer
3 Company Score Evaluation
The company score is computed as:
f (x) = xτw + b,
(9)
where w = ^П=1 αiyixi and b = 11 (x+ + x-)τw; x+ and x- are the obser-
vations from the opposite classes for which constraint (1) becomes equality.
By substituting the scalar product with a kernel function we will derive a
non-linear score function:
n
f(x) = K(xi , x)αi yi + b.
(10)
i=1
The non-parametric score function (10) does not have a compact closed
form representation. This necessitates the use of graphical tools for its visu-
alisation.
4 Variable Selection
In this section we describe the procedure and the graphical tools for selecting
the variables of the SVM model used in forecasts. We have two most im-
portant criteria of model accuracy: the accuracy ratio (AR), which will be
used here as a criterion for model selection, (Figure 6) and the percentage of
correctly classified out-of-sample observations. Higher values indicate better
model accuracy.
Fig. 6. The power curves for a perfect (green), random (red) and some real (blue)
classification models. The AR is the ratio of two areas A/B. It lies between 0 for a
random model with no predictive power and 1 for a perfect model.