A presentation illustrating the trade off between the true positive rate and the true negative rate of a classifier. Each point represents the classifier performance for a given threshold or ranking cut-off point.

See the full functionality:

Make your own chart:

The ROC Hull is the upper convex hull of the ROC chart. Each point represents the classifier performance for a given threshold or ranking cut-off point. Points on the ROC Hull represent an optimal performance of the classifier for certain misclassification costs.

See the full functionality:

Make your own chart:

The PR (precision recall) curve presents the trade off between the precision (the fraction of examples classified as positive that are truly positive) and the recall or true positive rate. Each point represents the classifier performance for a given threshold or ranking cut-off point.

See the full functionality:

Make your own chart:

The Lift curve plots the true positive rate (also found in ROC and PR curves) against the predicted positive rate (the fraction of examples, classified as positive). Each point represents the classifier performance for a given threshold or ranking cut-off point.

See the full functionality:

Make your own chart:

The Cost curve plots the normalized expected cost of the classifier as a function of the skew (fraction of positive examples multiplied by the cost of misclassifying a positive example) of the data on which it is deployed. Lines and points on the cost curve correspond to points and lines on the ROC curve of the classifier.

See the full functionality:

Make your own chart:

References:

Drummond and Holte (2006) - Cost curves: An improved method for visualizing classifier performance

The Kendall chart is defined as the difference between the normalized expected cost of the classifier and the normalized expected cost of an ideal classifier. Costs for both classifiers are calculated using the rate-driven threshold choice method.

See the full functionality:

Make your own chart:

References:

Hernández-Orallo, Flach and Ferri (2013) - ROC curves in cost space

The Rate-Driven chart is plots the expected loss for the classifier as a function of the skew (fraction of positive examples multiplied by the cost of misclassifying a positive example) of the data on which it is deployed. The cost is calculated using the rate-driven threshold choice method.

See the full functionality:

Make your own chart:

References:

Hernández-Orallo, Flach and Ferri (2013) - ROC curves in cost space