Abstract: Based on the traditional algorithm of density-based spatial clustering of applications with noise (DBSCAN), this paper determined the grouping effect by measuring cluster refinement service index. For groups with unsatisfactory clustering effect, the clustering parameters represented by scanning radius and minimum points were updated automatically according to their data characteristics, and then iterative clustering was carried out until the clustering effect was up to standard. At the same time, the DBSCAN algorithm was combined with the idea of node importance, which enabled it to output alternative sites. The results show that the improved DBSCAN algorithm can give the proper grouping according to the data characteristics, and the scanning radius and the minimum points parameters can be better adapted to the grouping situation, and the alternative nodes can effectively match the traffic resources around them.

Key words: clustering, DBSCAN algorithm, customized business bus, node importance, site location