Sequential Change-point Detection for High-dimensional and non-Euclidean Data
share
In many modern applications, high-dimensional/non-Euclidean data sequences are collected to study complicated phenomena over time and it is of scientific importance to detect anomaly events as the data are being collected. We studied a nonparametric framework that utilizes nearest neighbor information among the observations and can be applied to such sequences. We considered new test statistics under this framework that can make more positive detections and can detect anomaly events sooner than the existing test under many common scenarios with the false discovery rate controlled at the same level. Analytic formulas for approximate the average run lengths of the new approaches are derived to make them fast applicable to large datasets.
READ FULL TEXT
