MOSSFARM: Model structure selection by fuzzy association rule mining

Article type: Research Article

Authors: Pach, F.P. | Gyenesei, A. | Abonyi, J.

Affiliations: University of Pannonia, Department of Process Engineering, P.O. Box 158, 8200 Veszprem, Hungary | Department of Knowledge and Data Analysis, Unilever Research, Vlaardingen, P.O. Box 114, 3130 AC Vlaardingen, The Netherlands

Note: [] Corresponding author. Tel.: +36 88 624 209; Fax: +36 88 624 171; E-mail: abonyij@fmt.uni-pannon.hu

Abstract: Effective methods for feature and model structure selection are very important for data-driven modeling, data mining, and system identification tasks. This paper presents a new method for selecting important variables (regressors) in nonlinear (dynamic) models with mixed discrete (categorical, fuzzy) and continuous inputs and outputs. The proposed method applies fuzzy association rule mining. The selection process of the important variables is based on two interesting measures of the mined association rules.

Keywords: Feature selection, model order selection, fuzzy association rules, process modeling

Journal: Journal of Intelligent & Fuzzy Systems, vol. 19, no. 6, pp. 399-407, 2008