Finding optimal locations for taxi stands on city map

Abstract

We present an algorithm which suggests suitable locations for taxi stands on a city map. This is done using pickup locations information gathered from GPS devices installed on taxis. We aim at finding locations where taxi drivers are more likely to get a passenger or locations from where passengers can be reached by travelling minimum possible distance. To do the above mentioned tasks, we have modified CLARANS algorithm. The two major changes suggested by us include 1) Selecting initial seeds by dividing space using Quadtree approach 2) Swapping medoids with non-medoids in 'nearby' areas. We have tested our method using trajectory data generated over 90 days in Singapore and have compared our solution with the optimal solution obtained using PAM algorithm. We also show that not much variation occurs in clusters' quality if we increase the size of 'nearby' areas but running time increases to almost double the value.