ROC AUC
ELI5 - The Vibe Check
A ROC curve is what happens when you slide your model's how-sure-do-I-need-to-be-before-I-say-yes dial from paranoid to reckless and plot what happens to its hits and false alarms at every setting. The AUC is just one number that summarizes the whole curve: a perfect model scores 1, a model randomly guessing scores 0.5, so you can compare two models without staring at two squiggly lines.
Real Talk
The ROC (Receiver Operating Characteristic) curve plots the true positive rate against the false positive rate across every possible classification threshold, and the AUC (Area Under the Curve) condenses that curve into a single score between 0 and 1 that summarizes overall discriminative ability. An AUC of 0.5 means the model performs no better than random guessing, while 1.0 means perfect separation between classes.
When You'll Hear This
"The new model's AUC went from 0.81 to 0.89, that's a real improvement, not noise." / "Don't optimize for accuracy alone on an imbalanced dataset, look at the ROC AUC instead."
Related Terms
Class Imbalance
Class imbalance is when your fraud detector trains on data where 99.9% of transactions are totally normal, so it can hit 99.
Confusion Matrix
A confusion matrix is a little table that stops your model from lying to you with a single accuracy number.
F1 Score
The F1 Score is the balanced average of precision and recall — a single number that captures both.
Precision
Precision asks: 'Of all the times the AI said YES, how often was it actually right?
Recall
Recall asks: 'Of all the actual YES cases in the world, how many did the AI catch?' High recall means the model finds almost everything it should.