I'm pleased to share our (Nasiri, Miandoabchi, and Javadi) recently published article in the Group Decision and Negotiation entitled: Developing a Two-Stage Decision-Making Method for Selecting and Clustering Suppliers Based on the Resilience Criteria. As per the Springer Nature Content Sharing policy in getting personalized URL free access, those colleagues who are interested in reading and downloading this paper, can use the following link:
Abstract
The selection of appropriate suppliers is a critical issue for the survival of a company in a competitive market environment and is also one of the most significant challenges for organizations. The present study proposes a method for supplier selection by organizing them using clustering techniques. In this study, suppliers are selected based on a set of resilience criteria. The Improved Best Worst Method was used to determine the weight of the criteria using GAMS software. The two clustering algorithms including K-means and DBSCAN were used in this study. The DBSCAN algorithm was used to identify the noise points as the K-means algorithm could not identify these points properly. Both algorithms were implemented in the MATLAB software considering a scenario with 30 suppliers and 22 resilience criteria. The criteria including raw material quality, delivery time of raw materials, and reliability have the highest priority. Based on the results, some managerial implications were also presented.
Keywords: DBSCAN · K-means · Supplier Selection · Resilience · Clustering
This paper focuses on the clustering of suppliers based on resilience criteria, considering that some data (suppliers) may have noise, so this noise is identified and removed using the DBSCAN algorithm.
Conclusions and Future Study Recommendations
Many companies face difficulties in selecting suitable suppliers. Organization suppliers based on their characteristics can help to solve this problem. This study proposes a method for selecting suppliers by organizing them using clustering algorithms. The innovations of this research include the combination of methods used for weighting criteria and clustering suppliers. Two algorithms, including K-means and DBSCAN, were used for supplier clustering. The reason for using these two algorithms together was that the K-means algorithm cannot detect noise. Additionally, since an organization spends approximately 60–80% of its total sales price on purchasing raw materials and components, working with a supplier identified as noise can reduce the product quality and therefore demand, ultimately harming the entire SC and causing the organization to be eliminated from competition. For this reason, the DBSCAN algorithm was used in this study to identify the noise points. We fully examined a numerical problem consisting of 30 suppliers. First, we implemented the K-means algorithm, in which the optimal number of clusters was obtained as 2 based on the elbow method. Since the similarity and difference criterion is based on the Euclidean distance within and outside the cluster, the lower is the intra-cluster distance and the higher the inter-cluster distance, the better is the clustering quality. In the first step, when the K-means algorithm was implemented, the intra-cluster distance was 0.0028 for the first cluster and 0.0015 for the second cluster, while the inter-cluster distance was 0.0631. By implementing the DBSCAN algorithm, one supplier number in the first cluster was identified as noise and excluded from the set of suppliers. The K-means algorithm was then run again on the remaining suppliers. In this step, the intra-cluster distance decreased while the inter-cluster distance increased. Additionally, no noise was identified by running the DBSCAN algorithm and the algorithm was stopped after two iterations. This study led to the following conclusions:
● By removing the criteria with the highest weights, namely vulnerability, risk awareness, risk management culture, strategic risk planning and safety, the ranking of some suppliers changed. Therefore, it is recommended that managers consider these criteria when evaluating and selecting suppliers;
● Similarly, by removing the criterion of collaboration between SC members, which has the lowest weight, the ranking of suppliers changed. Therefore, it is recommended that managers also consider this criterion;
● Given that suppliers are at the top of the SC as a factor contributing to sustainability in a competitive environment, using the DBSCAN algorithm can help to identify noise suppliers (i.e. suppliers with a significant difference in performance compared to the other suppliers), which can provide a more homogeneous supplier base for the company.
It is obvious that the implementation of similar models (including the model of this study) in real applications requires master data, some of which are usually not available at the beginning of the project and create limitations in the implementation. In the case study of this article, there were some limitations in the estimation of the parameters. Having a suitable and powerful information system and a complete plan for collecting, updating and retrieving information such as a supplier portal requires the right amount of time and money. The implementation of supplier relationship management, in particular a clear agreement or contract between SC members, is also very useful.
Due to the several advantages of the elbow method, in this study this method is used to determine the value of K. The elbow method, while a useful tool for determining the optimal number of clusters in K-means clustering, has some drawbacks including subjectivity to the individuals analyzed, non-Gaussian data, sensitivity to initialization, and inefficient for large datasets. It is therefore recommended that this method is used in conjunction with another complementary method such as Silhouette.
According to the results of the current research, the following recommendations can be made for future research. First, BWM-I shows an ability to be successfully combined with other decision-making techniques. To further improve BWM-I, the current research can be developed in a grey fuzzy environment and the results can be compared with the current findings. Moreover, to determine the noise using the DBSCAN algorithm, after implementing the K-means algorithm, we examined the entire data and finally excluded the detected noise from the data set. The clusters obtained by the K-means algorithm can be examined separately to identify noise in future research. In addition, this research examined the clustering of suppliers. Future studies are recommended to cluster the customers so that manufacturers can ultimately produce products according to the needs and demands of the end users in each cluster. It is also suggested that more attention be paid to the environment issues by incorporating a number of more related green supplier selection criteria such as green production and environmental management system (e.g. the ability and flexibility to comply with the new environmental policies and regulations) to the present problem.