(TD)2PaM: A Constraint-Based Algorithm for Mining Temporal Patterns in Transactional Databases

Abstract

The analysis of frequent behaviors regarding temporal issues begins to achieve some interest, in particular in the area of health care. However, existing approaches tend to ignore the temporal information and only make use of the order among events occurrence. In this paper, we introduce the notion of temporal constraint, and propose three instantiations of it: complete cyclic temporal constraints, partial cyclic temporal constraints and timespan constraints. Additionally, we propose a new algorithm-(TD) 2 PaM, that together with these constraints, makes possible to focus the pattern mining process on looking for cyclic and timespan patterns. Experimental results reveal the algorithm to be as efficient as its predecessors, and able to discover more informed patterns.

A Comprehensive Survey of Pattern Mining: Challenges and Opportunities

“…Further, this framework is extended to incorporate time constraints deep into mining process and deal with complete periodicity. TD 2 PM standing for Temporality in Transactional Domain-Driven Pattern Mining [61] is the result of this extension which includes timespan and partial cyclic constraints. This algorithm allows for the discovery of patterns according to any time granularity chosen, without any further pre-processing.…”

Finding Cyclic Patterns on Sequential Data

“…On the other hand, [6] presents interesting results in terms of patterns' representation. In this approach cyclic patterns are represented based on their duration, period and the first instant where they occur, introducing additional information to the comprehension of its behavior once we can estimate how likely the pattern is to occur in future time instants.…”

“…Though, in next section, we present our approach for mining sequential cyclic patterns, bringing together the best properties of the work in [6] and [4].…”