Using clustering methods to analyze sales of auto parts at a truck service station

Abstract

Clustering methods are widely used to divide goods into groups depending on sales volumes in order to build an optimal purchasing planning and inventory management strategy. Cluster analysis methods do not provide an unambiguous partition of the original set of objects, therefore, in the work, existing clustering methods were analyzed to study sales of auto parts at truck service stations. To solve the problem, the following methods were chosen: k-means, hierarchical agglomerative clustering and DBSCAN. Before using the k-means method, the elbow method found the optimal number of clusters. The DBSCAN method is based on object density and automatically determines the number of clusters. The initial data for cluster analysis was information on sales of spare parts at truck service stations for 3 years; clustering was applied to data by year. The DBSCAN algorithm showed unsatisfactory results, since most of the goods (86%) were identified in one cluster, while others contained units of goods. The k-means method gave the best partitioning result, each group has a different volume. The distribution of goods in clusters changes over three years, so managers should study the change in the affiliation of goods to one group or another. The obtained clustering results will help determine the real needs of spare parts at truck service stations and build an optimal procurement strategy.