Hybrid data-driven approach for truck travel time imputation

Truck travel time data plays a critical role in freight performance measurement and is usually collected with probe-vehicle technologies. However, due to low sampling rates, truck data usually suffers from missing values. The primary purpose of this study is to develop a hybrid model to accurately impute missing truck travel time data by leveraging multiple data sources. The proposed model imputes missing values by considering the interaction, similarity, and differences of the data as well as incorporating available historical information. The hybrid model achieves robust results by combining both probe vehicle and loop detector data to impute continuous missing truck travel time data in sparse datasets. The proposed model was used to impute missing truck travel time data in the National Performance Measures Research Dataset (NPMRDS). The imputation performance of the proposed model was compared with several popular imputation models including historical, spline interpolation, random forest, and bootstrapping EM. The results indicated that the proposed model was capable of imputing missing data in sparse datasets, notably when the data was missing continuously. With similar to 13% mean-absolute percentage error, the hybrid model outperformed other models in imputing an entire day of missing data.