CPM: A general feature dependency pattern mining framework for contrast multivariate time series

•Unsupervised framework to mine contrast patterns in controlled experiment.•Customizable regularization techniques.•Efficient optimization algorithm easily adapt to various models under the framework.•Highly interpretable results in real world controlled experiments. With recent advances in sensor technology, multivariate time series data are becoming extremely large with sophisticated but insightful inter-variable dependency patterns. Mining contrast dependency patterns in controlled experiments can help quantify the differences between control and experimental time series, however, overwhelms practitioners’ capability. Existing methods suffer from determining whether the differences are caused by the intervention or by different states. We propose a novel Contrast Pattern Mining (CPM) framework to find the intervention-related differences by jointly determining and characterizing the dynamic states in both time series via multivariate Gaussian distributions. Under the CPM framework, we not only propose a new covariance-based contrast pattern model, but also integrate our previous proposed partial correlation-based model as a special case. An efficient generic algorithm is developed to optimize various CPM models by adjusting one of the sub-routines. Comprehensive experiments are conducted to analyze the effectiveness, scalability, utility, and interpretability of the proposed framework.