یک روش مبتنی بر انتخاب ورودی ANFIS و مدل سازی برای مشکل انتخاب تامین کننده

Supplier selection is a key task for firms, enabling them to achieve the objectives of a supply chain. Selecting a supplier is based on multiple conflicting factors, such as quality and cost, which are represented by a multi-criteria description of the problem. In this article, a new approach based on Adaptive Neuro-Fuzzy Inference System (ANFIS) is presented to overcome the supplier selection problem. First, criteria that are determined for the problem are reduced by applying ANFIS input selection method. Then, the ANFIS structure is built using data related to selected criteria and the output of the problem. The proposed method is illustrated by a case study in a textile firm. Finally, results obtained from the ANFIS approach we developed are compared with the results of the multiple regression method, demonstrating that the ANFIS method performed well.

Supplier selection plays an important role in the success of a company’s strategic goals. Changing customer preferences, public procurement regulations, and new organizational forms with more decision-makers make the purchasing function more complex and important for companies in today’s environment (De Boer, Labro, & Morlacchi, 2001). In addition, performing the purchasing function effectively and building strong and reliable partnerships with suppliers ensures that the company is more competitive in the market. An adequate method with appropriate selection criteria is necessary for a company to achieve a competitive advantage. In practice, supplier selection includes several tangible and intangible factors. Weber et al. reviewed and classified 74 articles which have appeared in the literature since 1966 (Weber, Current, & Benton, 1991). The study categorized these articles with respect to the 23 criteria of Dickson’s study. These criteria were originally based on a questionnaire sent to purchasing agents and managers from the United States and Canada. Dickson concluded that quality, delivery, and performance history are the three most important criteria. On the other hand, Weber and his colleagues noted that 47 of the 74 articles (64%) discussed more than one criterion. The two main articles that address the supplier selection criteria structure describe a multi-criteria view of the problem. The review of selection criteria based on various articles is shown in Table 1. Table 1. Supplier selection criteria research. Selection criteria A B C D E F G H I After sales service X X X Amount of past business X X X Attitude X Communication system X X Conflict resolution X Delivery X X X X X X Desire for business X Ease of communication X X X Economy X Financial position X X X X X X X Flexibility and response to changes X X X X X Geographical location X X Impression and skill X X Labor relations record X Management and organization X X X Operating controls X Packaging ability X Performance history X X Political stability X Price X X X X X Procedural compliance and discipline X X Production facilities and capacity X X X X X X X Quality X X X X X X X X Reciprocal arrangements X X Relationship closeness X X Reputation and position in industry X X Technical capability and technology X X X X X X X X Terrorism X Training aids X Warranties and claim policies X X A. Dickson (1966); B. Lee (2009); C. Haq and Kannan (2006); D. Chan and Kumar (2007); E. Chen et al. (2006); F. Liu and Hai (2005); G. Xia and Wu (2007); H. Ghodsypour and O’Brien (1998); I. Yahya and Kingsman (1999). Table options Several methods for supplier selection have appeared in literature, including approaches based on fuzzy logic. The main reason for a fuzzy logic approach is the need to handle vagueness and ambiguity in the problem. Researchers try to build effective models that not only consider quantitative aspects but also convert human judgments about qualitative criteria into meaningful results. For the first time in a fuzzy supplier selection problem, Amid et al. present an asymmetric approach that enables decision makers to assign different weight for each criterion (Amid, Ghodsypour, & OBrien, 2006). Their fuzzy multi-objective linear model has the capability to capture the fuzziness of the problem and order quantities can easily be assigned to each supplier under various constraints. Chen et al. presented a fuzzy TOPSIS approach by applying trapezoidal fuzzy numbers to assess the importance level of each criterion and ratings of alternative suppliers with regard to selected criteria (Chen, Lin, & Huang, 2006). In this model, a closeness coefficient is defined to determine the ranking order of all alternative suppliers by calculating the distances to fuzzy positive and negative ideal solutions (Chen, 2000). Chan and Kumar implemented a Fuzzy Extended Analytic Hierarchy Process (FEAHP) model that includes four hierarchies for a global supplier selection problem (Chan & Kumar, 2007). The study also discusses the risk factors related to a global view of the problem. Bevilacqua et al. integrated the fuzzy logic approach with a Quality Function Deployment method for a supplier selection problem in a medium to large industry that manufactures complete clutch couplings (Bevilacqua, Ciarapica, & Giacchetta, 2006). In this model, alternative suppliers are ranked according to their fuzzy suitable index values. Kwong et al. introduced a combined scoring method with a fuzzy expert systems approach to perform supplier assessment (Kwong, Ip, & Chan, 2002). In the case study, existing supplier assessment forms are used to assign the score of each individual supplier. Then obtained scores are used as inputs to build fuzzy if-then rules. Finally, the designed fuzzy expert system is implemented in the C programming language. In another study, Carrea and Mayorga applied a Fuzzy Inference System (FIS) approach to a supplier selection problem for new product development (Carrera & Mayorga, 2008). Their model includes 16 variables categorized in four groups and each group has an individual output. MATLAB FIS Editor is used to define rules and solve the problem. The proposed FIS system uses Gaussian and Bell membership functions to define the shape of both input and output variables. Ohdar et al. and Famuyiwa et al. also applied a Fuzzy Inference System approach to the supplier selection problem using the MATLAB FIS editor (Famuyiwa et al., 2008 and Ohdar and Ray, 2004). The main point of the Fuzzy Inference System approach is to determine fuzzy if-then rules from experts’ opinions. ANFIS, unlike FIS, automatically produces adequate rules with respect to input and output data, and takes advantage of the learning capability of neural networks. Many researchers and academicians concentrate on a fuzzy logic approach for the supplier selection problem, but not much attention is given to fuzzy logic with neural networks. Nassimbeni and Battain applied the ANFIS approach to evaluate the contribution that suppliers have on product development (Nassimbeni & Battain, 2003). The three inputs of model are product concept and functional design, product structural design and engineering, and process design and engineering. These inputs are used to evaluate suppliers and the sum of the weighted score of experts’ ratings corresponding to 15 selected criteria taken as output for the model. The data for 12 suppliers were used to instruct the neuro-fuzzy system, and data from the other four was used to test the results. In this article, output depends on subjective judgment of experts and focuses on supplier evaluation in a New Product Development (NPD) environment. We also discuss the topic of selecting a membership function type. There have been no prior applications of the neuro-fuzzy approach to the supplier selection problem and with respect to this fact a new model based on ANFIS is developed. For the first time in a supplier selection problem, ANFIS is used for both selection of criteria and developing the model of the problem. The model output is defined to be the share of each supplier’s sales. We also discuss selection of the number and type of membership functions. After the construction of the database, the model has two main stages: ANFIS input selection is executed first, and then the ANFIS model is built with respect to the related input/output data pattern. The paper is organized as follows: the next section introduces the basics of ANFIS. Section 3 includes a literature review of ANFIS. In Section 4, we present the algorithm for the model we developed. Section 5 includes a case study of the model. Finally, we present our conclusions in the last section.

In today’s global and competitive environment, firms should build an effective supplier base and select adequate partnerships by applying solid analytical techniques. In this paper, we present a new analytical technique, based on the ANFIS model, for supplier selection decision-making. After constructing the database, the model consists of two main stages: input selection with ANFIS, and building the ANFIS model using selected inputs from previous stage. To evaluate the efficiency of the model, MLR is applied to the same data; the ANFIS model we present outperformed the MLR according to the metrics of RMSE and MAPE. The ANFIS model we propose takes advantage of the learning capability of neural networks to build a useful analytic structure for decision-making related to supplier selection. From the point of view of a company, this structure can be easily applied to future purchasing decisions. To apply this strategy, decision makers from the company assign weightings for each assessment criteria for each alternative supplier. Then, the ANFIS system calculates output values that model the sales share of each supplier. Finally, the company can purchase items based on the obtained results. This decision support system facilitates the purchasing and decision making process of the company. This improved decision-making process provides a competitive advantage for the company, helping it to compete in the textile marketplace. The ANFIS model we have developed is robust with respect to the probable types of changes in the business. For example, if a new supplier enters consideration, or if the company decides to discontinue its relationship with an existing supplier, the ANFIS model will still work with the same criteria structure. On the other hand, if experts decide to incorporate a new criterion, the model loses efficiency because its method of producing results depends on application of historical data. If the company anticipates that the criteria structure will change in the future, the database structure should be constructed with consideration for these planned changes. For future work, other criteria reducing methods, such as clustering analysis, principal component analysis, linear discriminant analysis, and independent component analysis, may be applied with ANFIS for the supplier selection problem. However, the input selection phase we presented in this paper is also based on ANFIS. This is preferred because it presents a quick and straightforward method for input selection and a comprehensive solution that contains ANFIS in the implementation stages. If the problem includes multiple outputs, extended versions of ANFIS exist, such as Coactive neuro-fuzzy inference system (CANFIS) and multiple ANFIS (MANFIS). Finally, integration of ANFIS with linear programming may be considered as a topic for future research in problems with various constraints such as capacity and budgeting.