Profiling of Clusters of Activity-Travel Sequences Using a Genetic Algorithm

Classification of similar travel behavior is essential for market segmentation research in geography and transportation science. Cluster analysis using sequence alignment measurement incorporates the sequential information embedded in activity-travel sequences. The resultant clusters are then typically associated with the relevant variables. However, although the sequences are clustered by similar sequential information, the summary of the clusters do not reflect the sequential information with scientific rigor. This is because of the non-numeric characteristics of the sequential information. The study aims to develop a method for finding a representative sequence (RepSeq) that better profiles the cluster of sequences. The suggested method employs a genetic algorithm to search for a sequence potentially closest to the centroid by computing the smallest sum of distances from the searched sequence to all sequences of the cluster using a sequence alignment method. The suggested method is also applied to the real sequence data of the use of transport modes in Seoul. The result provides useful information for cluster interpretation and the subsequent analyses.