Enhanced imputation of GPS traces forcing full or partial consistency in activity-travel sequences : comparison of algorithms

Imputation of activity-travel sequences from global positioning system (GPS) data is dominantly based on epoch-level characteristic and land use data. This study shows that the accuracy of GPS data imputation can be improved by enforcing consistencies in transportation mode use across the day. In particular, this paper proposes a new algorithm to reduce misclassification error in imputed activity-travel diaries. The suggested algorithm identifies a hierarchical set of tours and superimposes logically consistent transportation modes at the tour-level. Three different methods, which keep some degree of consistency while allowing different transportation modes within a tour, are examined. Method I identifies the most probable transportation mode for each trip episode separately, and then compares the modes across all segments of the tour. Method II selects the most probable mode across the whole travel episode. Method III selects the most likely main mode in the intermediate portion of a tour as the mode in all trip segments. All methods impute transportation mode based on the highest number of epochs for which the predicted probability of that mode is highest. The algorithm was examined using GPS data recently collected in the Netherlands. Results show that the new algorithm significantly improves imputation accuracy of transportation modes. The enhanced algorithm, which partly relaxes the strict assumption of consistency, yields results even closer to reality