Event detection from time series data

摘要: In the past few years there has been increased interest in using data-mining techniques to extract interesting patterns from time series data generated by sensors monitoring temporally varying phenomenon. Most work assumed that raw is somehow processed generate a sequence of events, which then mined for episodes. some cases rule determining when sensor reading should an event well known. However, if phenomenon ill-understood, stating such difficult. Detection events environment focus this paper. Consider dynamic whose behavior changes enough over be considered qualitatively significant change. The problem we investigate identifying points at change occurs. statistics literature called change-point detection problem. standard approach (a) upriori determine number change-points are discovered, and (b) decide function will used curve fitting interval between successive change-points. paper generalize along both these dimensions. We propose iterative algorithm fits model segment, uses likelihood criterion segment partitioned further, i.e. it contains new changepoint. present algorithms batch incremental versions problem, evaluate their with synthetic real data. Finally, initial results comparing detected those people visual inspection.