Handling missing values in GPS surveys using survival analysis
GPS technology has made it possible to evaluate the performance of outdoor advertising campaigns in an objective manner. Given the GPS trajectories of a sample of test persons over several days, their passages with arbitrary poster campaigns can be calculated. However, inference is complicated by the early dropout of persons. Other than in most demonstrations of spatial data mining algorithms where the structure of the data sample is usually disregarded, poster performance measures such as reach and gross impressions evolve continuously over time and require non-intermittent observations. In this paper, we investigate the applicability of survival analysis to compensate for missing measurement days. We formalize the task of modeling the visit potential of geographic locations based on trajectory data as our variable of interest results from dispersed events in space-time. We perform experiments on the cities of Zurich and Bern simulating different dropout mechanisms and dropout rates and show the adequacy of the applied method. Our modeling technique is at present part of a business solution for the Swiss outdoor advertising branch and serves as pricing basis for the majority of Swiss poster locations.