A Change Detector for Prior Probabilities of Classes

The majority of current concept drift detectors focus on the results of a base classifier. But if there is a change in the data distribution or in the prior probability of the classes, these methods are unable to identify these types of change. This paper proposes Prior Probability Change Detection Method (PCDM), a method suited to identify changes in the prior probabilities of the classes. It works by associating traditional drift detection methods to analyze how the instances belonging to each class changes in time. Experiments in 24 artificial datasets of six generators indicate that PCDM presented the best results considering the sensitivity metric, the Matthews Correlation Coefficient, and the F1 score without losing any performance in the specificity metric.