Supplier evaluation and selection process has a critical role and significant impact on purchasing management in supply chain. It is also a complex multiple criteria decision making problem which is affected by several conflicting factors. Due to multiple criteria effects the evaluation and selection process, deciding which criteria have the most critical roles in decision making is a very important step for supplier selection, evaluation and particularly development. With this study, a hybridization of fuzzy c-means (FCM) and rough set theory (RST) techniques is proposed as a new solution for supplier selection, evaluation and development problem. First the vendors are clustered with FCM algorithm then the formed clusters are represented by their prototypes that are used for labeling the clusters. RST is used at the next step of modeling where we discover the primary features in other words the core evaluation criteria of the suppliers and extract the decision rules for characterizing the clusters. The obtained results show that the proposed method not only selects the best supplier(s), also clusters all of the vendors with respect to fuzzy similarity degrees, decides the most critical criteria for supplier evaluation and extracts the decision rules about data.
A supply chain management (SCM) is a system consists of three key parts, which are: the supply focuses on obtaining raw materials to manufacturing, the manufacturing focuses on converting obtained raw materials into finished products and the distribution focuses on reaching these finished products to customers through distributors, warehouses and retailers. Supply chain activities begin with customer orders and end with customer satisfactions. Selection of suppliers plays a critical role in an organization because it heavily contributes to the overall performance of a supply chain system. Assessing suppliers and selecting suitable ones among them a complex and critical decision making problem due to considering several criteria such as quality, cost, service, production lead time and environmental impact [19]. Eventually firms should select the most appropriate suppliers, because significant supplier selection reduces the purchasing cost and improves corporate competitiveness however inaccurate selection of supplier may lead to problems of finance and operation.
In the literature, there are several approaches as linear weighting methods, total cost approaches, mathematical programming techniques, statistical methods and artificial intelligence approaches have been proposed for supplier selection and evaluation process, they mostly locks onto ranking the suppliers and selecting the most appropriate suppliers. However in this paper, all the vendors are clustered by similarity degrees among them so not only the most appropriate supplier is determined but also the supplier categories and the membership degrees to them are determined. With this aspect, the proposed approach makes the decision making process much more flexible. Furthermore, in the literature the majority of researches have focused on supplier selection or evaluation or development separately, but despite that, we focus on an integrated flexible and efficient decision making model for supplier selection, evaluation and development.
The proposed model can cope better with uncertainty than conventional methods because it is designed to be more like human decision making functioning by its clustering, rule induction and feature extraction modules. In the clustering module, suppliers are clustered with a fuzzy clustering algorithm – fuzzy c means (FCM) – to evaluate their performance and similarities degrees. For every vendor, the membership degrees to different clusters are calculated. Unlike traditional hard clustering schemes, such as k-means, that assign each data point to a specific cluster, the FCM algorithm employs fuzzy partitioning such that each data point belongs to a cluster to some degree specified by a membership grade [11]. If data groups are well-separated, the hard clustering approach can be a natural solution. However, if the clusters are overlapped and some of data belong partially to several clusters, then fuzzy clustering is a natural way to describe this situation. In this case, the membership degree of a data object to a cluster is a value from the interval [0, 1].
In terms of supplier selection, supplier can be described by large scaled attributes, which can be represented with features in the view of machine learning. Indeed the weights of these attributes considered differently while we are evaluating suppliers. So some intelligent methods should be used to determine the most critical and important supplier attributes. The feature extraction methods in machine learning are used to find which attributes are more efficient and important in a clustering or classification model. In the rule induction and feature extraction module of the proposed model, decision rules for these clusters are defined then the most efficient criteria in decision making process are discovered by a feature selection method based on RST and these criteria considered as the most important features for further supplier development process. This extraction would be also very valuable for the supplier firms. They can try to improve these attributes primarily to be preferred at the next time. A case study is conducted to illustrate the proposed system. The system can also be easily implemented with different real supplier selection problems.
Although all of these proposed supplier selection models have useful and interesting principles, none of them is an integrated long-term relationship system as our system which presents the supplier evaluation rules due to fuzzy clusters and the attributes of suppliers due to their importance degrees. Therefore, there is a space for the development of new intelligent approaches toward effective support in the evaluation of suppliers, mainly for long-term relationships, characterized by the important supplier attributes. Beyond these, the classification methods must use the previous experiences to evaluate the performance levels of the available suppliers. However, if the supplier evaluation is realized due to a hard or fuzzy clustering algorithm, the past experiences are not required any more. This point is the one of the important reasons of selecting the FCM algorithm as a machine learning technique for this study. After the clusters are formed the second important point is to represent these clusters accurately. The rough set technique is generally used as a classifier and rule extractor in the literature. But it cannot be very sufficient to extract confident rules when it is used as a classifier method. On the other hand, rough set is a very robust rule and core attribute extractor. Therefore the rough set is applied on the clusters formed by FCM which is the most efficient fuzzy clustering algorithm. Briefly, this paper contributes to the state-of-art of the supplier selection problem, presenting a new and novel approach that integrates FCM and RST to construct a long-term relationship with suppliers.
The rest of the paper is organized as follows: Section 2 surveys relevant literature. Section 3 provides brief background knowledge about FCM and RST. The proposed system and obtained results from a sample supplier selection problem are presented in Section 4. The final section discusses the findings and concludes with a summary of this study and future directions.
Supplier evaluation and selection problems have been solved by several methods in literature such as linear weighting methods, total cost approaches, mathematical programming methods, statistical methods and AI methods. Through them, AI methods designed to be more like human judgment functioning and can cope better with the complexity.
This study presents an effective hybrid system by FCM and RST for solving supplier selection, evaluation and development problem. With the proposed system, the suppliers are clustered by FCM algorithm then as the result of the algorithm the fuzzy membership degrees of each vendor to these clusters are determined. Additionally, this system enables the decision makers to consider the most critical and important criteria of the suppliers which are effective at improving their own performances by using RST.
Supplier evaluation and selection involves ambiguous and imprecise appraisals by fuzzy nature. The proposed system is established to solve the supplier evaluation and selection problem in which all evaluation criteria ratings are taken into account separately for each supplier by fuzzy similarity degrees. In conventional methods; although, all evaluation criteria are rated separately, the categorization of the suppliers are done based only on an aggregated value, for example total weight point (TWP). The calculated TWP value is used to measure supply performance of the vendor. But two suppliers that have exactly the same TWP value can have totally different grades according to evaluation criteria. It means different suppliers that have the same TWP value can be dissimilar from the supply performance point of view. Oppositely, two vendors with different TWP can be in the same category of supply behavior. The aggregated TWP value can cause loss of all criteria's separated effects and failures in evaluation process [22]. At the proposed system all evaluation criteria ratings are regarded separately for each supplier to group them in a high degree of accuracy. So the system can adequately handle the imprecision of human judgment. This contribution of the proposed system is very important.
In the previous studies, the importance degree of evaluation criteria is neglected. The all evaluation criteria considered as they have an equal degree of importance. In this study the core evaluation criteria of the suppliers are determined by a feature selection method based on RST. The core evaluation criteria are the most important, critical an efficient features for supplier development. With the help of proposed system, decision makers are aware of which criteria are critical in supplier selection mainly. Additionally, the suppliers who are members of the cluster labeled as “Must improve their performance” or “Must be pruned” are aware of which of their features must be improved primarily. One of the new and useful contributions of this study is deciding the evaluation criteria which must be improved for a better evaluation performance.
Another contribution of the study is the decision rules which are used to characterize the supplier data and are provided to decision makers in an apparent form. According to these rules, the new supplier data can be grouped in one of the clusters without repeating the clustering process. These valid rules can be stored for future decisions about suppliers.
Consequently, the novel architecture of proposed system is very flexible and can be easily applied to other real supplier selection problems and also other multi-criteria decision making problems, such as personnel appraisal. Furthermore the system can be executed whatever how much the dataset size.
