Comparing Clustering Results of Specialized Functional Models and Standard Models through Comparative Analysis of Traffic Speed Differential Data

Abstract

Traffic flow and speed differences between cars are important factors that indicate the likelihood and danger of collisions. A vital part of intelligent transportation systems is discovering important locations to monitor and ticket speeding vehicles. To find these locations, we study data from a low-density city. Recent research in clustering includes fitting time series data to a set of basis functions forming functional data. An important step in determining if functional methods can be applied to real-world problems is comparing results in application. We compare the clustering performance of new methods developed for functional data clustering with robust non-functional methods. Using the original data and factors that may affect traffic that were not used in clustering, weekday, month, and speed limit, we determine that functional methods outperform non-functional methods at providing understandable and relevant clusters.

Faculty Mentor: Dr. Mohamad El-Hajj