Kukic, M., and Bierlaire, M.

Population synthesis at the level of households

Speaker: Kukic Marija

STRC 2021

September 14, 2021

Modern transportation science requires advanced demand models to predict the needs for the mobility of individuals and goods. In order to calibrate those models, we need data as an input. However, having in mind the data privacy constraints and the unavailability of that data, synthetically generated data is being used. Typically, generated data are either on the level of individuals or at the level of households. Although several different methodologies exist for accurately and efficiently generating synthetic population data at the level of the households, there are two main gaps. Firstly, in those approaches, the generation of individuals and their matching into households is done separately, through two sequential processes. Secondly, the state-of-the-art techniques might generate unrealistic observations due to high dependence on data and the lack of control within the generation process. This project aims to develop a methodology to integrate the generation of the agents and their matching into households in a one-step process. In this paper, we are presenting the first framework component for synthetic household imputation. By imputation, we imply the process of expanding the given dataset by adding synthetic people grouped into households using the information of a given individual. Another objective is to investigate the integration of real-world constraints and examine the amount of control we can embed within the generation process. The method is tested using census data from 2015 and mobile data from 2019 on the territory of Switzerland.