Early Time-Series Classification with Reliability Guarantee
USDOE National Nuclear Security Administration (NNSA), Hyrum S. Anderson, Sandia National Lab. (SNL-NM), Albuquerque, NM (United States), Nathan H. Parrish, Maya R. Gupta · 2012
We consider the early classification of (incomplete) time-series data given a complete timeseries training set.The early classification problem arises naturally when test sample data is collected over time, or when costs must be incurred to collect the data.For example, for missile defense, it is important to determine the target type long before it reaches its target.A practical goal is to assign a class label as soon as enough data is available to make a good decision.This objective is formalized through the notion of reliability-the probability that a label assigned to the early, incomplete data matches that assigned to the complete data, and we propose a method to classify incomplete data only if a user-specified reliability threshold is met.Our approach models the complete data as a random variable whose distribution is dependent on the current incomplete data and the training data.The method differs from standard strategies in that our focus is on determining the reliability of the early classification decision, not only the accuracy.Proposed methods are tested on a set of open-domain time-series datasets; where the goal is to classify the time-series as early as possible while still guaranteeing that the reliability threshold is met.