Back to the Future: Changing the Direction of Time to Discover Causality

In this paper we present the idea of using the direction of time to discover causality in temporal data. The Temporal Investigation Method for Enregistered Record Sequences (TIMERS), creates temporal classification rules from the input, and then measures the accuracy of the rules. It does so two times, each time assuming a different direction for time. The direction that results in rules with higher accuracy determines the nature of the relation. For causality, TIMERS assumes the natural direction of time, which assumes that events in the past can cause a target event in the present time. For the acausality test, TIMERS considers a backward flow of time, which assumes that events in the future can cause a target event in the present. There is a third alternative that TIMERS considers, and that is an instantaneous relation, where events in the present occur at the same time as a target event.

Authors and Affiliations

1. Department of Computer Science, University of Regina, Regina, Saskatchewan, Canada, S4S 0A2

   Kamran Karimi

Editors and Affiliations

1. Department of Computing and Information Science, College of Physical and Engineering Science, University of Guelph, Guelph, Ontario, Canada, N1G 2W1

   Yang Xiang

2. Dépt. Informatique-Génie Logiciel, Université Laval, Pavillon Pouliot, Ste-Foy, PQ, Canada, G1K 7P4

   Brahim Chaib-draa

Reprints and permissions

© 2003 Springer-Verlag Berlin Heidelberg

Policies and ethics