An Online Detection Method for Identifying True Change Points

Abstract

Time series data can be analysed using the singular spectrum analysis (SSA) technique by decomposing the data into its most basic elements, such as trends, recurrent patterns, and noise. Change point detection, a sequential application of SSA, is a procedure that uncovers the locations in time series data when sudden changes in its attributes occur. In an offline setting, the objective is to detect change points by analyzing the full dataset simultaneously. On the other hand, in an online (sequential) setting, the goal is to detect changes as fast as feasible using streaming data points. In this study, a novel online change point detection algorithm is proposed, which uniquely combines a sequential application of SSA with an integrated alarm mechanism. This combination allows for real time detection while explicitly differentiating between genuine changes (true alarms) and fake changes (false alarms). Several Experiments are performed in MATLAB and it is found that the window width parameter N is essential for detection procedure. The selection of the appropriate parameter is critical to guaranty accurate online change point detection. Since extremely small values of N cause false detections by mistaking outliers for change points, whereas exceptionally large values of N fre quently miss true change points. Thus, the best value of N for striking a balance between sensitivity and accuracy is either medium-sized or large.
The study concludes that the newly developed SSA-based algorithm han dles both synthetic and real-world data, offering a reliable tool for real time data analysis with a low number of false alarms and precise detection
of notable changes.