Abstract for: Integrating System Dynamics and Probabilistic Deep Learning: A Hybrid Framework for Uncertainty-Aware Decision Support

Modeling complex systems under uncertainty is challenging, especially in a dynamic environment where feedback loops, nonlinear interactions, and time delays regulate system behavior. Traditional System Dynamics (SD) models are powerful tools for simulating these systems and identifying their underlying behavior. However, they are limited in handling parameter uncertainty during the system evolution, which is imperative for building confidence in the model and its decisions. An integrated framework is proposed to enhance the SD model with Probabilistic Deep Learning (PDL) to handle parameter uncertainty. Unlike deterministic models, which produce point estimates and use fixed parameters over time, this novel hybrid approach combining SD and PDL uses SD for the dynamic simulation of the systems and PDL for its strength in learning probabilistic parameter distribution from data to refine dynamic key parameters. This paper makes several contributions: first, it presents a hybrid SD-PDL framework that combines the dynamic simulation capabilities of SD with PDL for learning parameter distributions. Second, it demonstrates the effectiveness of this approach to handling uncertainty in an epidemiological case study. Finally, it provides a way of applying this framework to various domains in policy scenario analysis, focusing on uncertainty quantification. Probabilistic Deep Learning