roclab: An R package for ROC-optimizing binary classification

초록

In imbalanced binary classification, maximizing the area under the receiver operating characteristic (ROC) curve is a widely used criterion. However, a unified software implementation for constructing binary classifiers based on this criterion has been lacking. The R package roclab provides tools for building such classifiers with various surrogate loss functions. In addition, flexible choices of regularization penalties and kernel functions are available to cover a wide range of classification problems. To this end, the package provides functions for model training, evaluation, and cross-validation. Because the approach is computationally intensive for large-scale data, scalable approximation options are also included to enable efficient model training. Thus, the package serves as a practical solution for binary classification on large-scale imbalanced datasets.

키워드