Accurate Evaluation of Quickest Changepoint Detectors via Non-parametric Survival Analysis
Researchers have developed new non-parametric estimators, KM-ARL and KM-ADD, for evaluating quickest changepoint detection (QCD) methods. These estimators address the limitations of traditional ARL and ADD metrics when dealing with finite and irregular sequence lengths, drawing an analogy to survival analysis. The proposed methods are shown to be asymptotically unbiased and practically useful for model selection, with accompanying Python code available for implementation. AI
IMPACT Introduces improved evaluation metrics for changepoint detection, enhancing the reliability of time-series analysis in AI applications.