Table 2.

Confusion Matrixes for out of sample classification of the decision trees shown in Figure 4, pessimistic tree built using only 80% of the data, Random Forests classifier and RUSBoost algorithm. For the high-risk population on the tree in Figure 4 A), 15% (4/26) of patients developed pneumonitis. On the other hand, for the high risk population of the tree on Figure a B) 22.73% (5/22) developed pneumonitis. The true performance will be in between these values and the pessimistic estimation. At this stage, the performance of the random forest is equivalent to the decision trees nevertheless, RUSBoost performs better than all the algorithms. True Positive = TP, False Negative = FN, False Positive = FP, True Negative = TN.

Figure 4 A) tree
Pneumonitis	4 TP	4 FN
No pneumonitis	22 FP	171 TN
Figure 4 B) tree
Pneumonitis	5 TP	3 FN
No pneumonitis	17 FP	176 TN
Pessimistic tree
Pneumonitis	2 TP	6 FN
No pneumonitis	45 FP	148 TN
Random Forests
Pneumonitis	5 TP	3 FN
No pneumonitis	16 FP	177 TN
RUSBoost
Pneumonitis	6 TP	2 FN
No pneumonitis	28 FP	165 TN