A service framework for learning, querying and monitoring multivariate time series

We propose a service framework for Multivariate Time Series Analytics (MTSA) that supports model definition, querying, parameter learning, model evaluation, monitoring, and decision recommendation. Our approach combines the strengths of both domain-knowledge-based and formal-learning-based approaches for maximizing utility over time series. More specifically, we identify multivariate time series parametric estimation problems, in which the objective function is dependent on the time points from which the parameters are learned. We propose an algorithm that guarantees to find the optimal time point(s), and we show that our approach produces results that are superior to those of the domain-knowledge-based approach and the logit regression model. We also develop MTSA data model and query language for the services of parameter learning, querying, and monitoring.