With reference to the average radio coverage of BTS and the required spatial scale in traffic analysis, the size of cells was set as 500 meters by 500 meters. All the BTSs in the same cell were replaced by one equivalent BTS coordinated at the cell’s centroid. Figure 3 Illustration of raster data structure. The calculation Sunitinib c-kit inhibitor of the four critical parameters and the transformation of BTSs’ geographical coordinates were described in Algorithm 2. Algorithm 2 Transformation of geographic coordinates. The city territory of Shanghai was covered by the raster with 245 rows and 348 columns. In the output of the algorithm, the 23,918 actual BTSs throughout Shanghai were reduced to 10,303 equivalent

BTSs. 3.2. Identification of Activity Points The original mobile phone data describes the individual’s virtual activities and provides the basic information of time, location, and

frequency. The synthesis and summarization of this basic information enable the inference of physical activities and the accessibility to the individual behavior patterns. In this study, the activity point was defined as the location at which a certain mobile subscriber continuously stayed for no less than 30 minutes. Activity points act as critical anchor points in people’s daily trajectories, incorporating home and workplace as two particular kinds of activity points. A set of activity points arranged in chronological order formed the activity chain of a certain mobile subscriber. The identification of activity points can be carried out as Algorithm 3. Algorithm 3 Identification of activity points. 3.3. Measurement of Spatial Interaction The macroscopic zonal interaction can be obtained

through the aggregate analysis of activity chains. In the existing models, the spatial interaction is analyzed based on the concept of trips. However, as for mobile phone data, the extraction of single trips from the continuous daily trajectories is not easily accessible. Though the particular data processing may contribute to the relatively accurate trip identification, the extra operation is doomed to lower the efficiency Batimastat of mass data mining. In this study, the novel approach for spatial interaction analysis was proposed based on frequent pattern mining. The correlations and associations between different areas were applied to measure the spatial interaction. Frequent pattern is item sets that appear in a dataset with frequency no less than a user-specified threshold. In this study, identities of areas acted as item sets; and each transaction was a sequence of area identities obtained from the activity chain of a certain mobile subscriber. Concretely speaking, let M = m1, m2,…, mN be an item set, where mi, i = 1,2,…, N, represent the identity of the ith area. With the specific mapping relation between areas and geographical coordinates, the activity chain A could be converted to a sequence of area identities AI.