Logistic regression classifiers are often deployed with a threshold rather than the provided classifier.bestCategory() method. This method picks the category with the highest conditional probability, which, in a 3-way classifier, can be just above one-third. This recipe will show you how to adjust classifier performance by explicitly controlling how the best category is determined.

This recipe will consider the 3-way case with the p, n, and o labels and work with the classifier produced by the Classifier-building life cycle recipe earlier in this chapter. The cross-validation evaluation produced is:

Category p
Recall: 0.64
Prec  : 0.57
Category n
Recall: 0.41
Prec  : 0.54
Category o
Recall: 0.81
Prec  : 0.81

We will run novel data to set thresholds.