Online change-point detection with kernels

Change-points in time series data are usually defined as the instants at which changes their properties occur. Detecting change-points is critical a number of applications diverse detecting credit card and insurance frauds, or intrusions into networks. Recently authors introduced an online kernel-based change-point detection method built upon direct estimation density ratio on consecutive intervals. This paper further investigates this algorithm, making improvements analyzing its behavior mean square sense, absence presence change point. These theoretical analyses validated with Monte Carlo simulations. The performance algorithm illustrated through experiments real-world compared to state art methodologies.