Synthetic Population Generation with Multilevel Controls: A Fitness‐Based Synthesis Approach and Validations

Abstract

The application of disaggregate models for predictions and policy evaluations requires as inputs detailed information on the socioeconomic characteristics of the population. The early procedure developed for population synthesis involved the generation of a joint multiway distribution of all attributes of interest using iterative proportional fitting (IPF). Recognizing its limitations, including the inability to deal with multilevel controls, several alternate methods have been proposed in the last few years. This article presents a methodology called the fitness‐based synthesis (FBS) that directly generates a list of households to match several multilevel controls without the need for determining a joint multiway distribution. The application and validation results demonstrate both the feasibility of the approach and its improved performance relative to the IPF and methods using fewer control tables. This article also presents a comprehensive validation of the synthetic populations against the true populations and thereby demonstrates the ability of the FBS method to generate the multidimensional correlations among the attributes. The number of iterations to terminations is found to be between one and three times the number of households to be synthesized. In sum, the FBS is an efficient and scalable methodology that is easy to implement and as such is a valuable tool for generating the detailed socioeconomic characteristics need for applying disaggregate travel‐demand forecasting models.

Method evaluation, parameterization, and result validation in unsupervised data mining: A critical survey

“…Social sciences, in particular, have developed a rich tradition of generating artificial data for the simple reason that large-scale real-life experiments are impossible, unethical or both. Wu et al (2018), when proposing to use existing macrolevel population data to generate artificial populations, cite a number of works from computational sociology that follow a similar approach (Barrett et al, 2009;Bisset et al, 2006;Ma & Srinivasan, 2015;Müller & Axhausen, 2011;Namazi-Rad, Mokhtarian, & Perez, 2014). Unfortunately, they seem not to have been aware of the influence that agent-based interactions have on those data.…”

Generating Realistic Synthetic Population Datasets

“…hierarchical IPF [13] and two-stage IPF [21], were also developed for generating synthetic population data for various research purposes such as land use and transportation microsimulation. Compared to the IPFbased approach, Ma and Srinivasan [11] proposed a fitness-based synthesis method to directly generate synthetic population, and Barthelemy and Toint [3] introduced a sample-free synthetic population generator by using the data at the most disaggregated level to define the joint distribution. Besides generating synthetic population with the combinational optimization based technique, Namazi-Rad et al [14] also projected dynamics over the synthetic population using a dynamic micro-simulation model.…”

Synthesizing Population for Microsimulation-based Integrated Transport Models Using Atlantic Canada Micro-data

“…The population synthesis methodology described in the present paper is comparable to work done by Srinivasan et al (2008) [12] in the sense that it produces a list of households to match several multilevel controls without the need for a joint multiway distribution. The Fitness-Based Synthesis (FBS) presented by Srinivasan et al [12,13] solves the particular problems of traditional population synthesis such as: zero cell problems, computational resources (memory) and nonintegers cell value in the joint-distribution tables [2]. Additionally, the adoption of Monte Carlo simulation in the procedure is waived due to the generation of integer fitness values.…”