Precision calculations
precision(x, ...)# S3 method for accuracy
precision(x, ...)
output depends on input and meaning of the function (the term precision
is highly polysemic)
an object from which precision is to be computed
generic functionality, not used
accuracy: Compute precision and goodness for accuracy curves, after Deutsch (1997),
using the accuracy curve obtained with accuracy(). This returns a named vector with
two values, one for precision and one for goodness.
Mean accuracy, precision and goodness were defined by Deutsch (1997)
for an accuracy curve \(\{(p_i, \pi_i), i=1,2, \ldots, I\}\), where \(\{p_i\}\)
are a sequence of nominal confidence of prediction intervals and each \(\pi_i\)
is the actual coverage of an interval with nominal confidence \(p_i\).
Out of these values, the mean accuracy (see mean.accuracy()) is computed as
$$ A = \int_{0}^{1} I\{(\pi_i-p_i)>0\} dp,$$
where the indicator \(I\{(\pi_i-p_i)>0\}\) is 1 if the condition is satisfied and
0 otherwise. Out of it, the area above the 1:1 bisector and under the accuracy
curve is the precision
\( P = 1-2\int_{0}^{1} (\pi_i-p_i)\cdot I\{(\pi_i-p_i)>0\} dp, \)
which only takes into account those points of the accuracy curve where \(\pi_i>p_i\).
To consider the whole curve, goodness can be used
$$G = 1-\int_{0}^{1} (\pi_i-p_i)\cdot (3\cdot I\{(\pi_i-p_i)>0\}-2) dp.$$
Other accuracy functions:
accuracy(),
mean.accuracy(),
plot.accuracy(),
validate(),
xvErrorMeasures.default(),
xvErrorMeasures()