Table 5 Pairwise model comparison using McNemar’s test for binary classification.
From: Trade-offs between machine learning and deep learning for mental illness detection on social media
Model 1 | Model 2 | Statistic | p-value | Winner | Corrected p-value | Significant |
|---|---|---|---|---|---|---|
SVM | lightGBM | 6.75 | 0.009 | SVM | 0.037 | Yes |
SVM | RandomForest | 9.73 | 0.001 | SVM | 0.009 | Yes |
SVM | LogisticRegression | 13.27 | <0.001 | SVM | <0.001 | Yes |
SVM | ALBERT | 107.57 | <0.001 | ALBERT | <0.001 | Yes |
SVM | GRU | 21.04 | <0.001 | GRU | <0.001 | Yes |
lightGBM | RandomForest | 0.44 | 0.508 | lightGBM | 1.000 | No |
lightGBM | LogisticRegression | 0.00 | 0.964 | Tie | 1.000 | No |
lightGBM | ALBERT | 159.51 | <0.001 | ALBERT | <0.001 | Yes |
lightGBM | GRU | 49.60 | <0.001 | GRU | <0.001 | Yes |
RandomForest | LogisticRegression | 0.26 | 0.672 | LogisticRegression | 1.000 | No |
RandomForest | ALBERT | 158.77 | <0.001 | ALBERT | <0.001 | Yes |
RandomForest | GRU | 54.86 | <0.001 | GRU | <0.001 | Yes |
LogisticRegression | ALBERT | 184.25 | <0.001 | ALBERT | <0.001 | Yes |
LogisticRegression | GRU | 43.95 | <0.001 | GRU | <0.001 | Yes |
ALBERT | GRU | 42.05 | <0.001 | ALBERT | <0.001 | Yes |
