Semiautomatic imputation of activity travel diaries : use of global positioning system traces, prompted recall, and context-sensitive learning algorithms

The new generation of dynamic activity-based models requires multiday or multiweek activity-travel data. Global Positioning System (GPS) tracers may be a powerful technology to collect such data, but previous applications of this technology to collect data of full activity travel patterns (not just time, route, and location) still required a substantial amount of manual data imputation and processing and hence are still demanding for both respondent and researcher. A semiautomatic data imputation system would be a major breakthrough and would involve less respondent burden. This paper reports and illustrates the design of a system called TraceAnnotator that processes multiday GPS traces semiautomatically. The process of imputing transportation modes, activity episodes, and other facets of activity travel patterns is based on a learning Bayesian belief network (BBN), which represents the multiple relationships between spatial, temporal, and other factors, including errors in the technology itself. Activity type is identified by fusing GPS data with geographic information system land use data and personalized land use data. Land use data are built during the data collection process using reverse geocoding and an Internet-based prompted recall survey, which also allows checking and correction of any imputation errors. The prompted recall data are used to update the conditional probabilities of the BBN. Consequently, that the system can learn over time implies that imputation accuracy will improve over time, reducing respondent and researcher burden. A pilot study is presented and potential improvements of the learning algorithm are discussed