Confidence-based early classification of multivariate time series with multiple interpretable rules

In the process of early classification, earliness and accuracy are two key indicators to evaluate the performance of classification, and early classification usually weaken its accuracy to some degree. Therefore, how to find a tradeoff between two conflict objectives is a challenging work. So far, there are just a few work touched the quality of early classification on univariate time series, and the confidence estimation for early classification on multivariate time series (MTS) is still an open issue. In this paper, we focus on interpretably classifying MTS examples as early as possible while guaranteeing the quality of the classification results. First, a fast method is proposed to mine interpretable and local rules from the MTS training data. Second, a valid measure is advanced to estimate the confidence of early classification on MTS examples. Finally, a strategy is designed to execute confident early classification to assume the classification confidence meets customers’ requirement. Experiment results on seven datasets show that the effectiveness and efficiency of our proposed algorithm for confident early classification on multivariate time series.