Fuzzy reasoning method based on voting techniques for building fuzzy systems

In this paper a novel fuzzy reasoning method based on the weighted maximum rule is proposed, which was initially inspired from the weighted majority vote algorithm and a fuzzy inference method using the maximum rule. This fuzzy reasoning method could be used for fuzzy if-then rules systems performing classification, control, and function approximation. The applications of the proposed fuzzy reasoning method bring two main effects: heuristic weights tuning of already generated fuzzy rules and establishments of new paradigms for training and testing fuzzy if-then rules systems. The heuristic weights tuning of already generated fuzzy rules is enabled through learning results of the testing, therefore the proposed fuzzy reasoning method can be regarded as a fuzzy rules tuning method. Based on the testing performance of individual rules, each fuzzy rule obtains a measure of accuracy, as the rule with the higher accuracy gets the higher weight. Each weight is then appended to the corresponding fuzzy rule for the future decision maldngs. The future decisions are made based on the weights as well as the other conventional information of fuzzy decision maldng. The weight tuning capability of the proposed method also allows the establishments of the new paradigms for training and testing the fuzzy rules systems. This construction of fuzzy rules systems employs comparatively more diversified processes in order to reduce numbers of decision making errors. The proposed fuzzy reasoning method and its new paradigms of building fuzzy rules systems are explained as well as examined with the classification problems. Experiments were carried out to prove the applicability of the method by using the numerical dasiairis datapsila set. The obtained results show the good properties of the proposed method.