What is ROC Curves?

Data StructureDatabaseData Mining

ROC stands for Receiver Operating Characteristic. ROC curves are a convenient visual tool for analyzing two classification models. ROC curves appears from signal detection theory that was produced during World War II for the search of radar images.

An ROC curve displays the trade-off among the true positive rate or sensitivity (proportion of positive tuples that are recognized) and the false-positive rate (proportion of negative tuples that are incorrectly recognized as positive) for a given model.

Given a two-class problem, it enables us to anticipate the trade-off between the rate at which the model can accurately identify ‘yes’ cases versus the rate at which it mistakenly recognizes ‘no’ cases as ‘yes’ for multiple “portions” of the test set. Some increase in the true positive rate appears at the value of an increase in the falsepositive rate. The area under the ROC curve is an assess of the accuracy of the model.

It can operate an ROC curve for a given classification model, M, the model should be able to return a probability or ranking for the predicted class of every test tuple. It is required to rank the test tuples in descending order, where the one the classifier thinks is generally belong to the positive or ‘yes’ class occurs at the top of the list.

Naive Bayesian and backpropagation classifiers are appropriate, whereas including decision tree classifiers, can simply be changed so as to return a class probability distribution for every prediction. The vertical axis of an ROC curve defines the true positive rate. The horizontal axis defines the false-positive rate. An ROC curve for M is plotted as follows.

It is beginning at the bottom left-hand corner (where the true positive rate and false-positive rate are both 0), it can test the actual class label of the tuple at the first of the list. If it is a true positive (i.e., a positive tuple that was correctly defined), then on the ROC curve, it can change up and plot a point.

It displays the ROC curves of two classification models. The plot also display a diagonal line where for each true positive of such a model, it is generally to encounter a false positive.

Therefore, the closer the ROC curve of a model is to the diagonal line, the less efficient the model. If the model is best, originally it can likely to encounter true positives as it can change down the ranked list. Therefore, the curve can move steeply up from zero. Later, as it can start to encounter fewer and fewer true positives, and progressively false positives, the curve cases off and becomes more horizontal.

Updated on 16-Feb-2022 11:53:36