Residence-Workplace Identification and Validation Based on Mobile Phone Data: A Case Study in a Large-Scale Urban Agglomeration in China

Residence-workplace identification is a fundamental task in mobile phone data analysis, but it faces certain challenges in sparse data processing and results validation because of the lack of ground-truth labels. Previous studies have generally relied on frequency-based methods for inference and trip-based metrics for validation, posing limitations in reliability and efficiency. This paper aims to fill this gap by developing a systematic approach that ranges from data error categorization and processing, feature relevance examination and parameter optimization, and the development of performance metrics considering both residence and workplace validation. For residence-workplace identification, we use a spatiotemporal closeness criterion to deal with the sparsity of data and develop effective dwelling time to enhance frequency-based methods, using one-month cellular signaling records from nine cities in the Yangtze River Delta urban agglomeration in China. For validation, we propose a residence-workplace pair metric based on the population-adjusted number of users, enabling more efficient evaluation of home and work locations than trip-based metrics. Results show that the mean absolute percentage errors (MAPEs) for the Nanjing and Shanghai cases are 5.04% and 8.46%, respectively. Adopted and verified in the large-scale urban agglomeration, the proposed method would be reliable for extracting residence and workplace from low-resolution mobile phone data and contributing to a more accurate identification of urban commuting patterns.