The ROC curve (Receiving Operator Characteristics) is commonly used to measure the performance of a classifier. It also introduces terms like Precision, Recall, AUC, which are present in most articles that deal with machine learning.
The ROC curve is also used to compare different classifiers. The higher a curve is, the larger the area under it and the fewer mistakes the classifier makes.
In a nutshell, the classifier returns a response that is either bad (-) or good (+). We can evaluate it because to build a classifier we always have a database containing the expected answers.
Additional Resources: