Imputing trip purposes for long-distance travel

摘要: Planning and policy analysis at the national, state inter-regional corridor levels depends on reliable information forecasts about long-distance travel. Emerging passive data collection technologies such as GPS, smartphones, social media provide opportunity for researchers practitioners to potentially supplement or replace traditional travel surveys. However, certain important trip information, purpose, mode, travelers’ socio-demographic characteristics, is missing from passively collected data. One promising solution this issue impute based supplementary (e.g., land use) advanced statistical mining algorithms. This paper develops machine learning methods, including decision tree meta-learning, estimate purposes passenger A dataset simulated 1995 American Travel Survey development validation of methods. The predictive accuracy proposed methods evaluated several scenarios varying with extent availability inputs. research design will not only a practically useful approach purpose imputation, but also generate valuable insights future Results show that imputation all available decreases 95 % two (business non-business) 77 four (business, personal business, visit, leisure). Based two-purpose scheme, algorithms when input used (a full-information model), 72 minimum model utilizes If traveler’s characteristics are (possibly through other models), 91 %.