City Research Online

Abstract

Association rule mining from time series has attracted considerable interest over the last years and various methods have been developed. Temporal rules between discovered episodes provide useful knowledge for the dynamics of the problem domain and the underlying data generating process. However, temporal rule mining has received little attention over the last years. In addition, the proposed methods suffer from two significant drawbacks. First the rules they produce are not robust enough with respect to noise. Second the proposed methods are highly dependent on the choice of the parameters since small perturbations on the parameters lead to significantly different results. In this paper we propose a frame-work to derive temporal rules from time series. Our approach is based on episode rule mining that discovers temporal rules from time series in the frequency domain using the discrete cosine transform. The rules are then translated to temporal relations between time series patterns of arbitrary length. Experimental results of the proposed framework are presented in the relevant section.

Publication Type:	Report
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Departments:	School of Science & Technology > Computer Science

Export

Downloads