Source-linked AI summary
Composite Monte Carlo Decision Making under High Uncertainty of Novel Coronavirus Epidemic Using Hybridized Deep Learning and Fuzzy Rule Induction
Simon James Fong, Gloria Li, Nilanjan Dey, Ruben Gonzalez Crespo, Enrique Herrera-Viedma
TL;DR
The paper addresses epidemic decision-making with incomplete, changing data and uncertain direct costs. It combines composite Monte Carlo simulation with optimized neural forecasting and fuzzy rule induction, producing narrower medical-cost ranges and interpretable epidemic-control rules.
Problem
Estimating epidemic behavior and virus-fighting costs is difficult because early decisions rely on limited data, unknown factors, and high uncertainty.
Method
The GROOMS+CMCM methodology combines deterministic forecasts selected and optimized by machine learning with probabilistic inputs, while FRI adds interpretable conditional rules to Monte Carlo outputs.
Results
The BFGS+PNN-based CMCM produced a narrower total_daily_cost range of 54mil–74mil and estimated fortnightly direct medical costs at about 73.6 million USD.
Takeaways & Limitations
The methodology complements macro-level probability distributions with interpretable fuzzy rules that indicate possible epidemic inflection outcomes and extreme future ranges.
Takeaways & Limitations
The FRI formulation assigns likelihood to whole rules rather than indicating each conditional test’s contribution, so the methodology supplements rule scoring with sensitivity-chart relations.
Abstract
from arXiv · showhide
In the advent of the novel coronavirus epidemic since December 2019, governments and authorities have been struggling to make critical decisions under high uncertainty at their best efforts. Composite Monte-Carlo (CMC) simulation is a forecasting method which extrapolates available data which are broken down from multiple correlated/casual micro-data sources into many possible future outcomes by drawing random samples from some probability distributions. For instance, the overall trend and propagation of the infested cases in China are influenced by the temporal-spatial data of the nearby cities around the Wuhan city (where the virus is originated from), in terms of the population density, travel mobility, medical resources such as hospital beds and the timeliness of quarantine control in each city etc. Hence a CMC is reliable only up to the closeness of the underlying statistical distribution of a CMC, that is supposed to represent the behaviour of the future events, and the correctness of the composite data relationships. In this paper, a case study of using CMC that is enhanced by deep learning network and fuzzy rule induction for gaining better stochastic insights about the epidemic development is experimented. Instead of applying simplistic and uniform assumptions for a MC which is a common practice, a deep learning-based CMC is used in conjunction of fuzzy rule induction techniques. As a result, decision makers are benefited from a better fitted MC outputs complemented by min-max rules that foretell about the extreme ranges of future possibilities with respect to the epidemic.
1. Introduction
The paper addresses epidemic decision-making when data are incomplete, dynamic, and uncertain, focusing on estimating resources needed to slow COVID-19 transmission. It proposes combining probabilistic Monte Carlo modeling with data-driven forecasting and interpretable fuzzy rules.
- COVID-19 response decisions require estimating sufficient resources despite incomplete knowledge of transmission and changing epidemic conditions.The paper frames this as a data analytics challenge for early intervention planning.
- Monte Carlo methods provide probabilistic information for analyzing possible epidemic outcomes and associated risks.Prior epidemic studies used Monte Carlo modeling because epidemic behavior and its effects involve substantial uncertainty.
- BFGS-PNN filters important variables and forecasts future time-series inputs before they enter the Monte Carlo simulator.The pre-processor addresses naive Monte Carlo's lack of input-variable importance selection.
- The proposed CMCM combines historical-data distributions with future predictions from an optimized deterministic forecasting model.This hybrid design allows different inputs to be represented according to their uncertainty.
- The methodology applies fuzzy rule induction to sensitivity-chart feedback, producing interpretable conditional rules with probabilities or certainty values.The rules complement Monte Carlo probability distributions by expressing decision conditions and consequences.
- The paper presents GROOMS+CMCM as a methodology that filters and integrates multiple data sources for epidemic forecasting and decision support.The paper states that the methodology includes BFGS-PNN for forecasting and FRI for generating fuzzy decision rules.
2. A Novel Methodology
The proposed GROOMS+CMCM methodology combines optimized neural-network preprocessing, composite Monte Carlo simulation, and fuzzy rule induction to model uncertain epidemic behavior. It complements probabilistic outputs with interpretable rules and min-max bounds for decision support.
- BFGS-PNN: BFGS-PNN preprocesses non-deterministic data by selecting salient features and forecasting inputs for the Monte Carlo model.
- BFGS-PNN: BFGS-PNN uses nonlinear, heuristic polynomial expansion and quasi-Newton optimization to minimize forecasting error.
- GROOMS+CMCM filters and condenses raw data from multiple sources into insights about future behaviors.
- Fuzzy Rule Induction: FRI infers interpretable conditional rules that complement Monte Carlo probability-density outputs and provide fuzzy upper and lower bounds for decisions.
- Fuzzy Rule Induction: A stated FRI drawback is that its rule-level likelihood does not indicate how each conditional attribute contributes to the outcome.The methodology proposes using sensitivity-chart scores and majority voting to address this limitation.
- Fuzzy Rule Induction: FRI assigns certainty and support indicators to rules, with the predicted class chosen by the greatest support value.
3. Experiment and Results
The experiment evaluates GROOMS+CMCM for forecasting COVID-19 medical costs from official data, combining probabilistic inputs, BFGS-PNN forecasts, sensitivity analysis, and fuzzy rules. BFGS-PNN produced narrower and more realistic cost forecasts than linear regression, while the resulting distributions and rules supported risk-based budgeting and epidemic-control interpretation.
- Experiment setup: The experiment used CDCP COVID-19 time-series data and selected infection, recovery, fatality, and cost-related variables for the CMC model.The simulation targeted direct medical costs for urgent national budget planning.
- Experiment setup: The simplified CMC model represented uncertain epidemic-control factors probabilistically and estimated costs from quarantine and isolation interventions.The authors present the model as an example that can scale beyond two direct inputs.
- Decision making under uncertainty: RMSE was approximately 128K with linear regression versus 62K with BFGS+PNN, whose forecasts produced a narrower total_daily_cost range of 54mil–74mil versus 12mil–83mil.The paper attributes the linear-regression result to over-forecasting non-stationary and upward-trending inputs.
- Decision making under uncertainty: At 50%, 80%, and 98% certainty, forecast mean budgets were respectively $74mil, $79mil, and $90mil, with different minimum and maximum ranges.The paper presents the 80% certainty option as a practical decision point, requiring about $79mil.
- Sensitivity chart and fuzzy rules: Sensitivity analysis ranked recovery duration in Days 10 and 12 and average isolation cost per day in Day 2 as the three most influential variables on total medical cost.The analysis links late recovery timing and early isolation costs to the final budget.
- Sensitivity chart and fuzzy rules: The fuzzy rules identified winning conditions involving declining new confirmed cases and failure conditions involving low cured rates with sustained high new confirmed cases.Rule 1 uses a three-day decline below 3581, while Rule 4 combines cured_rate below 3.86% with new daily confirmed cases between 1874 and 3350.
4. Conclusion
The paper addresses epidemic decision-making under incomplete, limited, and changing information by combining GROOMS with a composite Monte-Carlo model and fuzzy rule induction. This framework generates probabilistic outcome ranges and additional rule-based insights for decision support.
- The epidemic created urgent decision needs while information about its early development was scarce and incomplete.
- GROOMS+CMCM combines deterministic forecasting with probabilistic inputs sampled from data distributions.The method is designed to accommodate both deterministic and non-deterministic data within one Monte Carlo simulation.
- The composite Monte-Carlo model produces ranges of possible epidemic outcomes with associated probabilities for decision support.The resulting outputs can support sensitivity analysis, what-if analysis, and scenario planning.
- Fuzzy rule induction adds rule-based insights to the GROOMS+CMCM methodology.
Biography
The paper’s authors work across computer science, data analytics, collaborative computing, biomedical technology, and decision-support-related research. Their affiliations and roles span universities, laboratories, and academic leadership positions in several countries.
- Simon Fong is an Associate Professor at the University of Macau and an Adjunct Professor at Durban University of Technology.
- Gloria Tengyue Li is a PhD student at the University of Macau and leads a data analytics laboratory at the Zhuhai Institute of Advanced Technology.
- Nilanjan Dey is an Assistant Professor at Techno International New Town and serves as Editor-in-Chief of the International Journal of Ambient Computing and Intelligence.
- Rubén González Crespo is Vice Chancellor of Academic Affairs and Faculty at UNIR and Global Director of Engineering Schools for the PROEDUCA Group.
- Enrique Herrera-Viedma is a Computer Science and AI professor at the University of Granada and Vice-President for Research and Knowledge Transfer.