Adaptive Methods for Activity Monitoring of Streaming Data

Activity monitoring deals with monitoring data (usually streaming data) for interesting events. It has several appli- cations such as building an alarm or an alert system that triggers when outliers or change points are detected. We discuss desiderata for such a system. Then, assuming that the data can be modeled by linear models, we describe an adaptive incremental method for detecting outliers and change points in data streams. Our algorithm uses (a) in- tuitive criteria for labeling a data point as an outlier or as a change point, and (b) an adaptive incremental model es- timation method. In this paper, we use a forgetting factor- based Recursive Least Squares algorithm for adaptive in- cremental model estimation. We also present experiment results using both simulated and real data, which show that our algorithms for change and outlier detection could ac- curately detect these events.