Application of metaheuristic based fuzzy K-modes algorithm to supplier clustering

Many companies find difficulty in choosing right suppliers. Organizing suppliers based on their characteristics might help the company solve this problem. This study proposes an approach for supplier selection by organizing suppliers using a clustering method. Unlike other supplier segmentation methods, the proposed method analyzes suppliers' characteristics only based on the products they can offer, since this data is relatively easier to obtain. Furthermore a fuzzy K-modes clustering approach is applied to deal with overlapping classes. The reason is that some suppliers might have similar characteristics and belong to more than one class. Fuzzy clustering can allow this situation. Instead of using the original K-modes algorithm, this study proposes an improvement of fuzzy K-modes algorithm. Fuzzy K-modes algorithm is sensitive to the initial centroids. If the initial centroid is bad, it will not converge to a good clustering result. Therefore, this study combines fuzzy K-modes algorithm with a metaheuristic approach. Herein, the metaheuristic is responsible for giving more promising initial centroids for fuzzy K-modes algorithm. There are three metaheuristic approaches applied in this study, particle swarm optimization (PSO) algorithm, genetic algorithm (GA), and artificial bee colony (ABC) algorithm. The proposed metaheuristic-based fuzzy K-modes algorithms are verified using benchmark datasets before applying to the real supplier segmentation problem. The case study considers a supplier segmentation problem on automobile parts suppliers in Taiwan. The experiment results prove that metaheuristic-based fuzzy K-modes algorithm surpasses fuzzy K-modes algorithm. Between three tested metaheuristics, GA-based fuzzy K-modes algorithm is the most promising algorithm.