An Interactive Integrated MCDM Based on FANN and Application in the Selection of Logistic Center Location

Many practical problems are characterized as decision making with multiple, conflicting and noncommensurable nonlinear objectives and complex criteria. For identifying these nonlinear function structure and multiple attributes is a very difficult, time consuming, and confusing job in multiple criteria decision making (MCDM). This paper presents a feedforward artificial neural network (FANN) as a flexible and versatile approach for solving MCDM problems. In order to make the proposed ANN more adaptive to MCDM, this paper presents improvement methods to the FANN. Multicriteria hierarchy structure is established in an interactive method to determine appropriate network input-output structure. Fuzzy AHP is integrated with FANN. Criteria weights evaluated by Fuzzy AHP are used as ANN initial weights, which can make the FANN's weights get the definable meaning and avoid to be selected randomly. The method to standardize the criteria dimension can make the sigmoid activation function more efficient in ANN training. Arithmetic average function is applied to produce criteria value repeatedly to promise the generalization of ANN. We apply the proposed method to evaluate the location of logistic centers. The rank result indicates this model is a feasible and practical MCDM method which analyses both the qualitative and quantitative factors.