Ishita approach to construct an interval-valued triangular fuzzy regression model using a novel least-absolute based discrepancy

摘要

Regression models in the presence of fuzziness imposed researchers to construct regression models in three cases based on the fuzziness of the model's components - independent, dependent variables and the parameters. The purpose of this study is to construct full interval-valued triangular fuzzy regression model (IVTFRM) (when the regression components are represented as interval-valued triangular fuzzy numbers (IVTFNs). This paper proposes an approach (named as Ishita approach) using a mathematical linear optimization problem to find IVTFRM's parameters of the minimum discrepancy between the given patterns and predicted values of the dependent variable. For the same and behind the fact that the more uncertainty of an IVTFN is reduced, the more accuracy is gotten; the paper proposed a novel least absolute discrepancy between two IVTFNs which take into the consideration the distance between the maximum points of the piecewise uncertainty function of the two IVTFNs. Moreover, compared to the existing approach, the proposed model is general to fit any type of datasets whereas the existing model can only fit positive interval-valued triangular fuzzy data. The proposed approach is demonstrated with a real-life application. The model of proposed Ishita approach for the real-life application shows its superiority over the model of the only one existing approach.