Table 3.
The best performing results for each combination of feature types on the restricted dataset
Feature set | Test | wth | ucth | TP | FP | FN | TN | P | R | NPV | Spec | Acc | F1 |
Baseline | 17 | 6 | 27 | 186 | 73.90 | 38.60 | 87.32 | 96.90 | 86.00 | 50.70 | |||
concepts | χ2 | – | 300 | 37 | 15 | 7 | 177 | 71.15 | 84.09 | 96.20 | 92.19 | 90.68 | 77.08* |
concepts+assert | χ2 | – | 300 | 37 | 12 | 7 | 180 | 75.51 | 84.09 | 96.26 | 93.75 | 91.95 | 79.57** |
Words | t | 10 000 | – | 34 | 6 | 10 | 186 | 85.00 | 77.27 | 94.90 | 96.88 | 93.22 | 80.95** |
words+assert | χ2+t | 10 | – | 40 | 12 | 4 | 180 | 76.92 | 90.91 | 97.83 | 93.75 | 93.22 | 83.33** |
words+concepts | t | 10 000 | 50 | 37 | 8 | 7 | 184 | 82.22 | 84.09 | 96.34 | 95.83 | 93.64 | 83.15** |
words+concepts+assert | t | 10 000 | 5000 | 36 | 4 | 8 | 188 | 90.00 | 81.82 | 95.92 | 97.92 | 94.92 | 85.71** |
*p<0.01, **p<0.001; statistically significant differences in performance between the system configurations considered and the baseline.
Acc, accuracy; F1, F1-measure; FN, false negatives; FP, false positives; NPV, negative predictive value; P, precision; R, recall; Spec, specificity; TN, true negatives; TP, true positives.