A self-supervised neural-analytic method to predict the evolution of COVID-19 in Romania

Radu D. Stochitoiu, Traian Rebedea, Ionel Popescu, Marius Leordeanu

Abstract

Analysing and understanding the transmission and evolution of the COVID-19 pandemic is mandatory to be able to design the best social and medical policies, foresee their outcomes and deal with all the subsequent socio-economic effects. We address this important problem from a computational and machine learning perspective. More specifically, we want to statistically estimate all the relevant parameters for the new coronavirus COVID-19, such as the reproduction number, fatality rate or length of infectiousness period, based on Romanian patients, as well as be able to predict future outcomes. This endeavor is important, since it is well known that these factors vary across the globe, and might be dependent on many causes, including social, medical, age and genetic factors. We use a recently published improved version of SEIR, which is the classic, established model for infectious diseases. We want to infer all the parameters of the model, which govern the evolution of the pandemic in Romania, based on the only reliable, true measurement, which is the number of deaths. Once the model parameters are estimated, we are able to predict all the other relevant measures, such as the number of exposed and infectious people. To this end, we propose a self-supervised approach to train a deep convolutional network to guess the correct set of Modified-SEIR model parameters, given the observed number of daily fatalities. Then, we refine the solution with a stochastic coordinate descent approach. We compare our deep learning optimization scheme with the classic grid search approach and show great improvement in both computational time and prediction accuracy. We find an optimistic result in the case fatality rate for Romania which may be around 0.3% and we also demonstrate that our model is able to correctly predict the number of daily fatalities for up to three weeks in the future.

Introduction

Predicting the spread of an infectious disease like COVID-19 is notoriously difficult because underlying parameters—such as transmission rates, infectiousness periods, and fatality rates—vary significantly by region and over time. Traditional compartmental models like SEIR (Susceptible-Exposed-Infectious-Removed) are powerful, but fitting them to real-world data presents a complex "inverse problem." This problem is often non-convex, causing standard optimization techniques to get stuck in local optima. Furthermore, official "confirmed case" counts are often unreliable due to varying testing capacities and asymptomatic spread. This research addresses these challenges by using daily fatalities as the only "ground truth" anchor stable enough for accurate modeling.

The Modified-SEIR Framework

The researchers utilize a Modified-SEIR model that provides a more granular breakdown of the disease progression than the classic version. Instead of a single "Removed" category, the model tracks patients through various states of recovery and severity, including recovering at home with mild symptoms, recovering in the hospital with severe symptoms, and death. This system is governed by a set of Ordinary Differential Equations (ODEs).

The Modified-SEIR model flowchart illustrating the transitions between various disease states.

Neural-Analytic Optimization

To solve the inverse problem of mapping observed deaths back to model parameters, the authors propose a dual approach combining classical mathematics with deep learning:

  1. Self-Supervised Learning: A synthetic dataset of 100,000 fatality curves was generated by running the Modified-SEIR model with random parameter sets within defined ranges.
  2. The Neural Optimizer: A 1D Convolutional Neural Network (CNN) was trained to analyze sequences of daily deaths and predict the 11 parameters that generated them. This allows the network to learn the complex, non-linear relationship between disease dynamics and outcomes.
  3. Refinement: The CNN provides a near-instant "best guess" for real-world Romanian data, which is then fine-tuned using stochastic coordinate descent to minimize the difference between the model's prediction and actual reported deaths.
Parameters and initial values used in the Modified-SEIR model for the Romanian population.

Key Insights and Results

The neural-analytic approach proved vastly superior to traditional grid search in terms of computational efficiency. While a fine-grained grid search failed to find an optimal solution after 7 hours, the neural network provided a high-quality starting point almost instantaneously.

For the Romanian context, the model yielded several key epidemiological estimates:

  • Reproduction Number ($R_0$): Estimated at approximately 2.21.
  • Case Fatality Rate (CFR): Estimated between 0.245% and 0.3%, an "optimistic surprise" compared to many early global estimates.
  • Policy Impact: The model suggests that strict social distancing measures in Romania reduced the transmission rate by about 60%, successfully pushing the effective reproduction number below 1.0.
  • Predictive Accuracy: The model accurately predicted the peak of daily deaths (around April 18, 2020) and maintained low error rates (under 4%) for three-week-ahead forecasts.

Conclusion

This work demonstrates that focusing on fatalities as a reliable metric, combined with a self-supervised neural optimizer, can effectively overcome the noise and uncertainty inherent in pandemic data. While the model is sensitive to initial conditions like the pandemic start date, it provides a robust framework for forecasting and policy evaluation. Future work may involve extending the deterministic ODE system to Stochastic Differential Equations (SDEs) to better account for reporting noise and daily variability.

Summary updated

Abstract

Analysing and understanding the transmission and evolution of the COVID-19 pandemic is mandatory to be able to design the best social and medical policies, foresee their outcomes and deal with all the subsequent socio-economic effects. We address this important problem from a computational and machine learning perspective. More specifically, we want to statistically estimate all the relevant parameters for the new coronavirus COVID-19, such as the reproduction number, fatality rate or length of infectiousness period, based on Romanian patients, as well as be able to predict future outcomes. This endeavor is important, since it is well known that these factors vary across the globe, and might be dependent on many causes, including social, medical, age and genetic factors. We use a recently published improved version of SEIR, which is the classic, established model for infectious diseases. We want to infer all the parameters of the model, which govern the evolution of the pandemic in Romania, based on the only reliable, true measurement, which is the number of deaths. Once the model parameters are estimated, we are able to predict all the other relevant measures, such as the number of exposed and infectious people. To this end, we propose a self-supervised approach to train a deep convolutional network to guess the correct set of Modified-SEIR model parameters, given the observed number of daily fatalities. Then, we refine the solution with a stochastic coordinate descent approach. We compare our deep learning optimization scheme with the classic grid search approach and show great improvement in both computational time and prediction accuracy. We find an optimistic result in the case fatality rate for Romania which may be around 0.3% and we also demonstrate that our model is able to correctly predict the number of daily fatalities for up to three weeks in the future.
BibTeX
@misc{stochitoiu2020self,
  author = {Radu D. Stochitoiu and Traian Rebedea and Ionel Popescu and Marius Leordeanu},
  title = {{A self-supervised neural-analytic method to predict the evolution of COVID-19 in Romania}},
  year = {2020},
  eprint = {2006.12926},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2006.12926},
}
Figure