Compute support & confidence per rule (not just per ruleset), use that to 'clean' the ruleset according to a threshold.
Not so prioritary at the moment, because changing the score function to roc_auc_score instead of f1-score, or balanced accuracy, gave much better results and already relatively clean rules.
linomp
commented
2 years ago
Compute support & confidence per rule (not just per ruleset), use that to 'clean' the ruleset according to a threshold.
Not so prioritary at the moment, because changing the score function to roc_auc_score instead of f1-score, or balanced accuracy, gave much better results and already relatively clean rules.