What is the difference between ROC AUC and accuracy in model evaluation?

What is the difference between ROC AUC and accuracy in model evaluation?

ROC AUC (Receiver Operating Characteristic Area Under the Curve) measures the ability of a model to distinguish between classes, while accuracy simply measures the proportion of correct predictions.

When should I use ROC AUC instead of accuracy?

Use ROC AUC when you have imbalanced classes or when you want to evaluate the model's performance across different thresholds. Accuracy can be misleading in such cases.

Can you give an example of when accuracy might be misleading?

Sure! In a dataset with 95% negative and 5% positive samples, a model predicting all negatives would have 95% accuracy, but it wouldn't be useful at all.

How do I calculate ROC AUC?

You can calculate ROC AUC using libraries like scikit-learn in Python, which provides functions to compute the ROC curve and the AUC value.