Constraint learning using adaptive neural-fuzzy inference system

Purpose – The purpose of this paper is to present a new method for solving parametric programming problems; a new scheme of constraints fuzzification. In the proposed approach, constraints are learned based on deductive learning. Design/methodology/approach – Adaptive neural-fuzzy inference system (ANFIS) is used for constraint learning by generating input and output membership functions and suitable fuzzy rules. Findings – The experimental results show the ability of the proposed approach to model the set of constraints and solve parametric programming. Some notes in the proposed method are clustering of similar constraints, constraints generalization and converting crisp set of constraints to a trained system with fuzzy output. Finally, this idea for modeling of constraint in the support vector machine (SVM) classifier is used and shows that this approach can obtain a soft margin in the SVM. Originality/value – Properties of the new scheme such as global view of constraints, constraints generalization, clustering of similar constraints, creation of real fuzzy constraints, study of constraint strength and increasing the degree of importance to constraints are different aspects of the proposed method.