Using R libraries to facilitate sensitivity analysis and to calibrate system dynamics models

Evaluating the sustainability of complex health system transformation in the context of population ageing: An empirical system dynamics study

“…For that reason, we have found it more efficient to implement the model and conduct the numerical experimentations on a programming language platform, which in this case is Python 2.7.11. Although the common practice is to use purpose specific software, the use of programming languages has been recently introduced in the field of system dynamics (Duggan, 2018;Duggan, 2019;Dural-Selcuk et al, 2019). Our choice of implementation platform provided us with the flexibility to work with a higher level of granularity and the opportunity to demonstrate that it is feasible to work with a large number of stock disaggregation within the SD modelling paradigm.…”

A content analysis-based approach to explore simulation verification and identify its current challenges

“…Additional exploration is required in the presence of correlated inputs to distinguish the distribution of effect between correlated factors [113]. Numerous M&S studies utilize SA to facilitate simulation space and solution space exploration (for example, see Duggan [114], Thiele, Kurth [115], and Hekimoğlu and Barlas [116]).…”

Exploring the opportunity of using machine learning to support the system dynamics method: Comment on the paper by Edali and Yücel

“…Within this iterative model‐building process, there is scope for a wide range of tools and methods to support the process, beyond the ‘orthodox’ stock and flow models and causal loop diagrams. Methodological advances in areas such as machine learning and artificial intelligence can also provide system dynamics modellers with unprecedented opportunities to apply robust computational methods to model building (Duggan, 2019; Sterman, 2018). These methods and tools can be applied at different stages of the model‐building process, for example, SEIR model calibration of infectious disease outbreaks using Markov chain Monte Carlo (MCMC) simulation (Ghaffarzadegan & Rahmandad, 2020) and machine learning approaches to analyse model output (Kwakkel, 2017), and the paper by Edali and Yücel (2020) is a further example of progress in this area.…”

“…However, a possible barrier to entry for the use of machine learning methods within the system dynamics modelling process is the perceived complexity of the area and often the focus on big data problems involving real‐time analysis, for example, image processing. While there are machine learning problems that require highly computation approaches such as deep learning (Sengupta et al, 2020), many machine learning methods can be usefully applied to simulation data and explored using open‐source libraries such as R (Duggan, 2018, 2019).…”

“…As an illustrative example of how a machine learning method can be applied to time series data, and by extension, output from a system dynamics model, consider the following example using the R language (Duggan, 2018(Duggan, , 2019. It is based on a weather and energy data set for the Republic of Ireland (Duggan, 2020), which contains hourly weather observations for 2017, which include rainfall (rain), mean sea-level pressure (msl) and predominant wind direction in degree (wwdir).…”