Bayesian networks for spatial learning: a workflow on using limited survey data for intelligent learning in spatial agent-based models

Abstract

Machine learning (ML) algorithms steer agent decisions in agent-based models (ABMs), serving as a vehicle for implementing behaviour changes during simulation runs. However, when training an ML algorithm, obtaining large sets of micro-level human behaviour data is often problematic. Information on human behaviour is often collected via surveys of relatively small sample sizes. This paper presents a methodology for training a learning algorithm to guide agent behaviour in a spatial ABM using a limited survey data sample. We apply different implementation strategies using survey data and Bayesian networks (BNs). By being grounded in probabilistic directed graphical models, BNs stand out among other learning algorithms in that they can be based on expert knowledge and/or known datasets. This paper presents four alternative implementations of data-driven BNs to support agent decisions in a spatial ABM. We differentiate between training BNs prior to, or during the simulation runs, using only survey data or a combination of survey data and expert knowledge. The four different implementations are then illustrated using a spatial ABM of cholera diffusion for Kumasi, Ghana. The results indicate that a balance between expert knowledge and survey data provides the best control over the learning process of the agents and produces the most realistic agent behaviour.

Risk perception and behavioral change during epidemics: Comparing models of individual and collective learning

“…To explore the implications of intelligent learning on the gradient from individual to collective, we advance the existing cholera ABM (CABM) originally developed to study cholera diffusion [35]. In CABM, MLs steer agents' behavior [23,35,36], helping them to adjust risk perception and coping during an epidemic outbreak. For this study, we ran eight ABMs to test various combinations of individual and group learning, using different information sources-with or without interactions among agents-as factors in the BNs.…”

“…S1 Appendix provides further details on how the BNs are implemented, together with tables of the parameters. Sensitivity analysis of the aggregated model dynamics on the BNs inputs and training alternatives can be found in [23,36].…”

“…The participants-primarily students from developing countrieswere introduced to the problem of cholera disease, saw pictures of water, and were asked if they would use the water as it is (D1 in Fig 1), walk to a cleaner water point (D2), or use the water after boiling it (D3). The survey data were used to construct and train our BNs [36]. We also used these data to evaluate the results of expert-driven BNs in CABM [43].…”

Guest editorial for spatial agent-based models: current practices and future trends

“…[13]). How to represent and simulate behaviour in agentbased models was also a recurrent issue with two papers discussing how approaches borrowed from machine learning can be used to improve the representation of behaviour [14,15]. How to create models that could scale from the micro to macro was another theme with the point being made that current agent-based modelling architectures do not foster models that are easily translatable to a regional or global context (e.g.…”

“…Methods from machine learning are beginning to gain traction within agent-based modelling community, but these approaches are computationally intensive and often require large amounts of data. The potential benefits of machine learning for helping to simulate behavioural change is recognised by Abdulkareem et al [15]. They identify that one of the issues with accurately simulating behaviour is the large number of micro-level data sets that are required.…”