Credit scoring with class imbalance data: An out-of-sample and out-of-time perspective

Data for predicting default risk in the next 12, 36, and 60 months. A training sample is provided for each outcome window, as well as out-of-sample (OOS) and out-of-time (OOT) testing samples. OOS is generated using the same timeline as the training sample, whereas OOT is created from a future timeline. The samples are all generated using the same characteristics/features.