Data censorship involves two sets of corresponding values: known observations and latent (i.e., unknown) true values. Modeling of censored data has been researched in multiple works, including the famous Tobit model for known and deterministic censorship. However, when modeling demand for transport services, the censorship challenge becomes two-fold: not only is demand data inherently censored by limited supply, but it also typically lacks any account of the difference between observed and true demand. To address this problem, we devise and analyze two complementary methods for censored modeling, when no indication is given about the extent of censorship. The first modeling method is generic, while the other method is non-parametric and utilizes domain knowledge. Our experiments demonstrate both the importance of accounting for censorship and the capability of each method in reconstructing the underlying, latent patterns.
|Number of pages||10|
|Publication status||Published - 2021|
|Event||hEART 2020: 9th Symposium of the European Association for Research in Transportation - Lyon, France|
Duration: 3 Feb 2021 → 4 Feb 2021
|Conference||hEART 2020: 9th Symposium of the European Association for Research in Transportation|
|Period||03/02/2021 → 04/02/2021|