Admissible generalisation of temporal sequences as chronicles

Abstract

Machine learning is the art of generalising a set of examples. Beside the efficiency of the algorithms, the challenge is to define generalisations that make sense for a data scientist. In this article, we consider generalisations of temporal sequences as chronicles. A chronicle is a temporal model that represents a situation occurring in temporal sequences, i.e. a series of event types with timestamps. A chronicle is a collection of event types with metric temporal constraints on their delays of occurrence. Generalising sequences by a set of event types can intuitively be the smallest set of events that occur in all sequences. A question arises with the generalisation of metric temporal constraints. In this article, we study the admissibility of these generalisations by deriving the notion of rule admissibility to the generalisation as chronicles. Through formalisation, new insights about the notions of chronicles may lead to conceive original chronicle mining algorithms.