Travel Pattern Analysis from Trajectories Based on Hierarchical Classification of Stays

Abstract

GIScience 2016 Short Paper Proceedings Travel Pattern Analysis from Trajectories Based on Hierarchical Classification of Stays Ryo Inoue, Motohide Tsukahara Graduate School of Information Sciences, Tohoku University, 6-6-06 Aramaki Aoba, Aoba, Sendai 980-8579, Japan Email: {rinoue, tsukahara}@plan.civil.tohoku.ac.jp Abstract It has recently become possible to utilize a large amount of detailed trajectories for travel pattern analysis. However, methods proposed in this area have limitations in applying the data taken over a wide area. To overcome these limitations, this paper proposes extraction of travel patterns through hierarchical classifications of stays and travel patterns based on the Huffman coding algorithm. The results of experiments conducted using the proposed method on trajectories in Okinawa, Japan confirm its feasibility for analyzing travel patterns. 1. Introduction A huge amount of trajectory data has become available with the widespread use of positioning technology. One of its applications is tourist activity analysis, with research focused on extracting typical travel patterns actively underway. Previous studies in this area first detect and code stays, then analyze travel patterns from stay sequences. Many approaches have been proposed for the latter process. Zheng et al. (2007) analyzed transition probabilities between sites, Giannotti et al. (2007) mined sequential patterns, and Shoval and Isaacson (2007) and Shoval et al. (2015) used sequence alignment. However, the former process has not been sufficiently investigated. Most of the studies conducted simply judged stays by the preset area classification. Giannotti et al. (2007) and Zhang et al. (2009) extracted stays via density-based analyses; however, no threshold setting criteria were presented. Previous analyses are limited especially when the study area is wide. However, because places of interest and arrival and departure times vary, the number of travel patterns is significant. This results in difficulty finding similar patterns. Thus, adjusting the resolution of the analysis to reduce patterns is essential. This study focuses on “stays,” which each consists of a visited place and its arrival and departure time. The analysis resolution of a stay may vary from site- to region-basis in space and from minute- to day-basis in time. Analysis methods whose results change according to resolution settings are useless as they cause difficulty interpreting results. We believe that hierarchical classification of travel patterns to spatio-temporal resolution settings is key to solving the problem. Thus, this paper proposes classification of travel patterns based on Huffman coding of stays, and reports on tests of its applicability using trajectory data obtained in Okinawa, Japan. 2. Hierarchical Classification of Stays and Travel Patterns Huffman coding outputs compact code with average code length close to Shannon entropy: the average information contained in the data. It constructs a binary tree by repeating the aggregation