Hierarchical Neyman-Pearson Classification for Ordered Classes
Base Classifier Training function Train a base multi-class model (RF /...
HNP Box Plot Experiment
Delta search
Classes Mapping function for HNP Algorithm Map class labels to canonic...
hnp_summary Summarize a ternary classifier's performance
HNP Umbrella (flex): use custom score functions and pre-split data
HNP Umbrella Algorithm
Upper Bound of the ith Threshold (Optimal ith Threshold)
T1 Calculation Create T1 scoring function from a fitted model
T2 Calculation Create T2 scoring function as ratio P(class 2)/P(class ...
The Hierarchical Neyman-Pearson (H-NP) classification framework extends the Neyman-Pearson classification paradigm to multi-class settings where classes have a natural priority ordering. This is particularly useful for classification in unbalanced dataset, for example, disease severity classification, where under-classification errors (misclassifying patients into less severe categories) are more consequential than other misclassifications. The package implements H-NP umbrella algorithms that controls under-classification errors under user specified control levels with high probability. It supports the creation of H-NP classifiers using scoring functions based on built-in classification methods (including logistic regression, support vector machines, and random forests), as well as user-trained scoring functions. For theoretical details, please refer to Lijia Wang, Y. X. Rachel Wang, Jingyi Jessica Li & Xin Tong (2024) <doi:10.1080/01621459.2023.2270657>.