Using spatio-temporal data for estimating missing cycling counts: a multiple imputation approach

Abstract

A data-driven, yet novel, multiple imputation model was proposed to estimate missing counts at permanent count stations. The model was motivated by the spatial–temporal relationship of cycling volumes of nearby facilities as well as the strong correlation between cycling volumes and weather conditions. The proposed model is flexible as it assumes no prior knowledge about which locations may experience sensor malfunction (i.e. missing counts). As well, the model does not assume any prior knowledge of the relationship structure between the input variables. The model was tested using a large dataset of more than 12,000 daily bicycle volumes collected between 2009 and 2011 at 22 different count stations in the City of Vancouver, Canada. The model showed a strong estimation power with an average error of about 12.7%. Sensitivity analyses were carried out to investigate the impact of different model parameters on the estimation accuracy.