Classification of typical Bluetooth OD matrices based on structural similarity of travel patterns: Case study on Brisbane city

The structure of a daily Origin-Destination (OD) matrix represents the distribution of travel patterns in terms of number of trips ending into different destinations within a region. However, the daily travel patterns could be significantly different due to different characteristics such as regular working days, weekends, long weekends, public holidays, school holidays and special event days etc. Most of the travel patterns are recurrent in nature and they can be classified into different clusters of typical travel patterns represented by their corresponding typical OD matrices. Among many statistical measures, Structural SIMilarity (SSIM) index is identified as an appropriate statistical measure to classify the typical daily OD matrices based on the similarity of travel patterns. The paper discusses the strengths and practical limitations of state-of-the-art application of SSIM for structural comparison of OD matrices of large scale networks and proposes a new practical approach based on geographical window for using SSIM in transport applications. The SSIM is then used as a proximity measure for clustering that provides basis for the identification of typical daily OD matrices. The proposed approach is tested by a case study on a real Bluetooth based proxy OD matrices from Brisbane city, Australia.