A1134
Title: Adapted-threshold classification and monitoring: Boosting sensitivity
Authors: Ansgar Steland - RWTH Aachen University (Germany) [presenting]
Abstract: Many procedures for monitoring and classification thresholds are standardized statistics using constants to raise the alarm and mark suspicious data points. The proposal is to use threshold functions depending on covariates related to the alarm event. This allows for a boost in sensitivity for certain regions of the sample space, e.g., those corresponding to high risks. New results on optimal threshold functions controlling the marginal type I error rate are presented. Further, nonparametric estimation is studied for the proportional threshold. Lastly, the potential for improvement of classifiers from statistics and machine learning is discussed. The approach is illustrated by real data examples.
