A Geo-Probabilistic Model for Estimation of Call Detail Records Neighbourhood Allocation
ISBN 978-85-88783-11-9
Authors
1Arias Martins, P.; 2Bittencourt e Silva, P.
1PREFEITURA DA CIDADE DO RIO DE JANEIRO Email: pedroariasmartins@gmail.com
2COPPE - UFRJ Email: pedrobittencourt87@gmail.com
Abstract
This work presents a methodological approach on how to allocate CDR (Call Detail Record) information, naturally associated with the Cell Phone Antenna location, to the neighbouring areas to attend specific analytical demands (usually to allocate to city zones and/or neighbourhoods).The result is a linear operator to convert antenna allocated calls to zone allocated calls. The main issue related to geographical analysis of CDR data is that all calls from a given region are clustered on the closest (or serviced) antenna geographical position, generating severe distortions. It is known that the the exact position of the cell phone is impossible to reconstruct based only on CDR, but our approach allocate the call on the most probable area surrounding the antenna range of service. By this method, it is possible to estimate the proportional population of each Census Tract polygon that makes calls in the specific area of an antenna. The needed data to perform this process is common public data, except the CDRs themselves: Census Tract Polygons (CTP) geographical data with population, Cell Phone Towers coordinates, Street map and City limits polygons The methodology uses ESRI’s software ArcGis 10.1 to generate the Voronoi Polygons (VP) for each antenna, based on the Delaunay Triangulation method. This is a proxy approach for antenna range, since it does not considers terrain relief, eventual differences on particular antenna frequency, power and buildings interferences. A intersect geo-operation between the CTP polygons (that are contained in the city limits) and the VP is run, to determine the topological relations between them. Once it is determined, the areas of the resulted polygons (CTP’) is calculated absolutely and proportionally to each VP polygon, in order to have the proportion of each part on the upper VP. Afterwards, the CDR are added to the model, accordingly to the goals of the analysis being made. Which means that it makes it possible to allocate each call to an area based on a defined model of triangulation, and to estimate the population of each sub-area on the corresponded antenna. Formally, given a VP from an antenna i Vi, that intersects with n zone polygons {B1,B2,...,Bn} the first step is to calculate the proportional population of each intersection zone of Vi with Bj, Zij. Being P(.) the population and A(.) the area: P(Zij) =A(Zij)A(Bj)P(Bj)