Early and Revocable Time Series Classification

Abstract

Many approaches have been proposed for early classification of time series in light of its significance in a wide range of applications including healthcare, transportation and finance. Until now, the early classification problem has been dealt with by considering only irrevocable decisions. This paper introduces a new problem called early and revocable time series classification, where the decision maker can revoke its earlier decisions based on the new available measurements. In order to formalize and tackle this problem, we propose a new cost-based framework and derive two new approaches from it. The first approach does not consider explicitly the cost of changing decision, while the second one does. Extensive experiments are conducted to evaluate these approaches on a large benchmark of real datasets. The empirical results obtained convincingly show (i) that the ability of revoking decisions significantly improves performance over the irrevocable regime, and (ii) that taking into account the cost of changing decision brings even better results in general.