Time-series Change Point Detection with Self-Supervised Contrastive Predictive Coding

Change Point Detection (CPD) methods identify changes in the trends and properties of time series data in order to describe the underlying behaviour of the system. Detecting changes and anomalies in the web services, the trend of application usage, and sensor data can provide valuable insights into the system. We propose TS-CP2 a novel self-supervised approach for CPD that is based upon contrastive representation learning with a Temporal Convolutional Network (TCN). TS-CP2 is the first CPD approach to employ a contrastive learning strategy. Through extensive evaluations, we demonstrate that our method outperforms five different state-of-the-art CPD methods, including those adopting either unsupervised or semi-supervised approach. TS-CP2 is shown to improve both non-Deep learning- and Deep learning-based methods by 0.28 and0.12 in terms of average F1-score across three datasets, respectively.