Finding Temporal Patterns Using Constraints on

When the evolution of variables over time is relevant to a classification task, established classifiers cannot be applied directly as the typical input format (data table) is not appropriate. We propose a new representation of temporal patterns that includes constraints on (partial) presence, (partial) absence as well as the duration of temporal predicates. A general-to-specific search-based algorithm is presented to derive classification rules. The approach evaluates promising on artificial and real data.