Detecting and understanding urban changes through decomposing the numbers of visitors' arrivals using human mobility data

In recent years, mobility data from smart cards, mobile phones and sensors have become increasingly available. However, they often lack some of the key information including the purposes of trips for each individual user. Information on trip purposes is crucial for projecting the future travel patterns as well as understanding the characteristics of each area of a city and how it is changing. This paper proposes a method called EAT-CD (Extraction of Activity Types and Change Detection). It estimates the volume of passengers by activity types (e.g. commuting, leisure) using non-negative matrix factorization and detects changes in the number of visitors for each activity (e.g. increase in shopping trips triggered by the development of a new commercial facility). Validity of EAT-CD is tested through empirical analysis using smart card data of public transportation in Western Japan. The results showed that EAT-CD is effective in deriving activity patterns, which showed strong correlation with travel survey data. The results also confirmed that EAT-CD detects changes in travel patterns (e.g. start and end of semesters) and land uses (e.g. establishment of new facilities).