Extending Geographic Weights of Evidence Models for Use in Location Based Services

This dissertation addresses the use and modeling of spatio-temporal data for the purposes of providing applications for location based services. One of the major issues in dealing with spatio-temporal data for location based services is the availability and sparseness of such data. Other than the hardware costs associated with collecting movement related data, privacy issues are also a concern. A common approach adopted for behavior monitoring, while providing subscribed users with efficient location based services involves the use of historical movement trajectories. These approaches also involve the modeling of the users’ attribute characteristics via methods that include surveys, records of buying habits and web2.0 applications. Such approaches assume the availability of movement trajectory data and associated subject identifying databases. This could conflict with the issues of privacy and with the availability of such data (at a reasonable cost). The dissertation proposes methods that could be used for delineating physical space where movement occurs, as probability surfaces using geographic feature ‘evidences’. The dissertation builds upon weights of evidence models (Bonham-Carter et al. 1988) to deal with the sparseness of spatio-temporal data. The proposed methods allow location based service applications to be predictive in identifying relative regions of movement, and are able to deal with the scale of the analysis in addition to being able to model both, spatial and temporal variables together. This allows the generation of spatio-temporal predictions and maps of the probability surfaces. The methods presented here provide a mechanism for analytics and visualization that can be used by location based service providers. The study uses George Mason University campus as a hypothetical example for presenting and testing these methods using a simulated movement dataset of vector point locations and grid geographic features.