Implementing Bayesian network and generalized raking multilevel IPF for constructing population synthesis in megacities

Abstract

Constructing agent data with detailed information of their sociodemographics is substantially important for agent-based modelling. However, to collect whole population data is not efficient because it requires an expensive and time-consuming survey, especially for a large populations. Therefore, this research tries to construct the whole population of the Greater Jakarta area using previous surveys. One of them was conducted by JICA in 2009 with a 3% sample of households. This paper uses graphical representation, a Bayesian Network (BN), which allows identifying the best joint probability distribution of the data structure and Iterative Proportional Fitting (IPF) to fit data against aggregate census data. The results show that using BN approach can produce data that represents probability distribution of sample data and IPF to match it against aggregate census data.