Boosting Interval-Based Literals: Variable Length and Early Classification

In previous works, a system for supervised time series classification has been presented. It is based on boosting very simple classifiers: only one literal. The used predicates are based on temporal intervals. There are two types of predicates: i) relative predicates, such as “increases” and “stays”, and ii) region predicates, such as “always” and “sometime”, which operate ver regions in the domain of the variable. This work presents two new features of the system. First, the system can now deal directly with variable length time series. Second, the obtained classifiers can now be used with partial time series. This “early classification” is essential in some supervision environments, where it is necessary to give an alarm signal as soon as possible. Experiments on different data sets repositories, show that the proposed method is highly competitive with previous approaches.