Inverse problem for parameters identification in a modified SIRD epidemic model using ensemble neural networks

BioData Mining, Journal Year: 2023, Volume and Issue: 16(1)

Published: July 18, 2023

In this paper, we propose a parameter identification methodology of the SIRD model, an extension classical SIR that considers deceased as separate category. addition, our model includes one which is ratio between real total number infected and were documented in official statistics. Due to many factors, like governmental decisions, several variants circulating, opening closing schools, typical assumption parameters stay constant for long periods time not realistic. Thus objective create method works short time. scope, approach estimation relying on previous 7 days data then use identified make predictions. To perform average ensemble neural networks. Each network constructed based database built by solving days, with random parameters. way, networks learn from solution model. Lastly get estimates Covid19 Romania illustrate predictions different time, 10 up 45 deaths. The main goal was apply analysis COVID-19 evolution Romania, but also exemplified other countries Hungary, Czech Republic Poland similar results. results are backed theorem guarantees can recover reported data. We believe be used general tool dealing term infectious diseases or compartmental models.

Language: Английский

---

A Machine Learning-enabled SIR Model for Adaptive and Dynamic Forecasting of COVID-19

medRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: July 31, 2024

Abstract The COVID-19 pandemic has posed significant challenges to public health systems worldwide, necessitating accurate and adaptable forecasting models manage mitigate its impacts. This study presents a novel framework based on Machine Learning-enabled Susceptible-Infected-Recovered (ML-SIR) model with time-varying parameters predict dynamics across multiple geographies. incorporates emergent patterns from reported time-series data estimate new hospitalisations, hospitalised patients, deaths. Our adapts the evolving nature of by dynamically adjusting infection rate parameter over time using Fourier series capture oscillating in data. approach improves upon traditional SIR models, which often fail account for complex shifting due variants, changing interventions, varying levels immunity. Validation was conducted historical United States, Italy, Kingdom, Canada, Japan. model’s performance evaluated Mean Absolute Percentage Error (MAPE) Cumulative values (CAPE) three-month forecast horizons. Results indicated that achieved an average MAPE 32.5% 34.4% 34.8% deaths, forecasts. Notably, demonstrated superior accuracy compared existing like-for-like disease metrics, countries proposed ML-SIR offers robust tool dynamics, capable geographical contexts. adaptability makes it suitable localised hospital capacity planning, scenario modelling, application other respiratory infectious diseases similar transmission such as influenza RSV. By providing reliable forecasts, supports informed decision-making resource allocation, enhancing preparedness response efforts.

Language: Английский

---

LSTM-based Forecasting using Policy Stringency and Time-varying Parameters of the SIR Model for COVID-19

Published: March 3, 2023

Accurate forecasting of the number infections is an important task that can allow health care decision makers to allocate medical resources efficiently during a pandemic. Two approaches have been combined, stochastic model by Vega et al. for modelling infectious disease and Long Short-Term Memory using COVID-19 data government's policies. In proposed model, LSTM functions as nonlinear adaptive filter modify outputs SIR more accurate forecasts one four weeks in future. Our outperforms most models among CDC United States data. We also applied on Canadian from two provinces, Saskatchewan Ontario where it performs with low mean absolute percentage error.

Language: Английский

---

COVID-19 EDA analysis and prediction using SIR and SEIR models

International Journal of Healthcare Management, Journal Year: 2022, Volume and Issue: unknown, P. 1 - 16

Published: Oct. 25, 2022

Under the epidemic of emerging infectious diseases like new coronavirus disease 2019 (COVID-19), data analysis and scenario using mathematical models are important evidence that forms core policy decisions. Become, it is a research field has attracted more attention due to global COVID-19. In this paper, in addition epidemiological findings COVID-19, describe basic way thinking symptom model. We show here comparing other how SIR SEIR effective. model, we measures social distancing, vaccination have affected virus, can see reduced rate virus with few taken. impact been taking place comparison 40% 75% vaccinated people. spread decreased when number people increased. Here, conclude by checking whereas model accurate compared

Language: Английский

---

A Novel Approach on Deep Learning—Based Decision Support System Applying Multiple Output LSTM-Autoencoder: Focusing on Identifying Variations by PHSMs’ Effect over COVID-19 Pandemic

International Journal of Environmental Research and Public Health, Journal Year: 2022, Volume and Issue: 19(11), P. 6763 - 6763

Published: June 1, 2022

Following the outbreak of COVID-19 pandemic, continued emergence major variant viruses has caused enormous damage worldwide by generating social and economic ripple effects, importance PHSMs (Public Health Social Measures) is being highlighted to cope with this severe situation. Accordingly, there also been an increase in research related a decision support system based on simulation approaches used as basis for PHSMs. However, previous studies showed limitations impeding utilization policy establishment implementation, such failure reflect changes effectiveness restriction short-term forecasts. Therefore, study proposes LSTM-Autoencoder-based establishing implementing To overcome existing studies, proposed methodology predicting number daily confirmed cases over multiple periods output strategies rapidly identifying varies effects anomaly detection. It was that demonstrated excellent performance compared models time series analysis statistical deep learning models. In addition, we endeavored usability suggesting transfer learning-based can efficiently variations effects. Finally, provides multi-period forecasts, reflects variation policies. intended provide reasonable realistic information implementation and, through this, yield expected be highly useful, which had not provided systems presented studies.

Language: Английский

---

On the Effects of Type II Left Censoring in Stable and Chaotic Compartmental Models for Infectious Diseases: Do Small Sample Estimates Survive Censoring?

Proceedings of the AAAI Symposium Series, Journal Year: 2024, Volume and Issue: 2(1), P. 467 - 474

Published: Jan. 22, 2024

In this paper, we discuss a selection of tools from dynamical systems and order statistics, which are most often utilized separately, combine them into an algorithm to estimate the parameters mathematical models for infectious diseases in case small sample sizes left censoring, is relevant rapidly evolving remote populations. The proposed method relies on analogy between survival functions dynamics susceptible compartment SIR-type models, both monotone decreasing time determined by dual variable: hazard function prediction number infected people models. We illustrate methodology continuous model presence noisy measurements with different distributions (Normal, Poisson, Negative Binomial) discrete model, reminiscent Ricker map, admits chaotic dynamics. This estimation procedure shows stable results experiments based popular benchmark dataset samples. manuscript illustrates how classical theoretical statistical methods can be merged interesting ways study problems ranging more fundamental situations complex disease potential that applied large covariates types censored data.

Language: Английский

---

Long-Term Prediction of Large-Scale and Sporadic COVID-19 Epidemics Induced by the Original Strain in China Based on the Improved Nonautonomous Delayed Susceptible-Infected-Recovered-Dead and Susceptible-Infected-Removed Models

Journal of Computational and Nonlinear Dynamics, Journal Year: 2024, Volume and Issue: 19(4)

Published: Feb. 12, 2024

Abstract The COVID-19 virus emerged abruptly in early 2020 and disseminated swiftly, resulting a substantial impact on public health. This paper aims to forecast the evolution of large-scale sporadic outbreaks, stemming from original strain, within context stringent quarantine measures China. In order accomplish our objective, we introduce time-delay factor into conventional susceptible-infected-removed/susceptible-infected-recovered-dead (SIR/SIRD) model. nonautonomous delayed SIRD model, finite difference method is employed determine that transmission rate epidemic area exhibits an approximately exponential decay, cure demonstrates linear increase, death piecewise constant with downward trend. We employ improved SIR model for regions characterized by extremely low or nearly zero mortality rates. these regions, estimated through two-stage decay function variable coefficients, while removal aligns recovery previously mentioned results this study demonstrate high level concordance actual COVID-19, predictive precision can be consistently maintained margin 3%. From perspective parameters, it observed under strict isolation policies, China relatively has been significantly reduced. suggests government intervention had positive effect prevention country. Moreover, successfully utilized outbreaks caused SARS 2003 outbreak induced Omicron 2022, showcasing its broad applicability efficacy. enables prompt implementation allocation medical resources different ultimately contributing mitigation economic social losses.

Language: Английский

---

A Comprehensive Exploration of Mathematical Models and Machine Learning Techniques for COVID-19

Advances in bioinformatics and biomedical engineering book series, Journal Year: 2024, Volume and Issue: unknown, P. 1 - 27

Published: June 7, 2024

Mathematical modeling has proved to be useful in predicting the spread of infectious diseases and assessing dynamical behavior contagious diseases, including COVID-19. Various models aid forecasting COVID-19 spread, such as SEIR (Susceptible – Exposed Infected Recovered), SIR SIRD Recovered Death), SIRVD Vaccinated Death). With recent technological advancements, can also done using machine learning techniques SVM (support vector machine), decision tree, random forest, linear regression. This chapter delves into various mathematical provides simulations Python for These provide essential insights evaluate which algorithm performs better evaluation metrics.

Language: Английский

---

A Markov Switching Autoregressive Model with Time-Varying Parameters

Forecasting, Journal Year: 2024, Volume and Issue: 6(3), P. 568 - 590

Published: July 29, 2024

This study showcased the Markov switching autoregressive model with time-varying parameters (MSAR-TVP) for modeling nonlinear time series structural changes. enhances MSAR framework by allowing dynamic parameter adjustments over time. Parameter estimation uses maximum likelihood (MLE) enhanced Kim filter, which integrates Kalman Hamilton and collapsing, further refined Nelder–Mead optimization technique. The was evaluated using U.S. real gross national product (GNP) data in both in-sample out-of-sample contexts, as well an extended dataset to demonstrate its forecasting effectiveness. results show that MSAR-TVP improves accuracy, outperforming traditional GNP. It consistently excels error metrics, achieving lower mean absolute percentage (MAPE) (MAE) values, indicating superior predictive precision. demonstrated robustness accuracy predicting future economic trends, confirming utility various applications. These findings have significant implications sustainable growth, highlighting importance of advanced models informed policy strategic planning.

Language: Английский

---

From Policy to Prediction: Assessing Forecasting Accuracy in an Integrated Framework with Machine Learning and Disease Models

Journal of Computational Biology, Journal Year: 2024, Volume and Issue: 31(11), P. 1104 - 1117

Published: Aug. 2, 2024

To improve the forecasting accuracy of spread infectious diseases, a hybrid model was recently introduced where commonly assumed constant disease transmission rate actively estimated from enforced mitigating policy data by machine learning (ML) and then fed to an extended susceptible-infected-recovered forecast number infected cases. Testing only one ML model, that is, gradient boosting (GBM), work left open whether other models would perform better. Here, we compared GBMs, linear regressions, k-nearest neighbors, Bayesian networks (BNs) in COVID-19-infected cases United States Canadian provinces based on indices future 35 days. There no significant difference mean absolute percentage errors these over combined dataset [ H(3)=3.10,p=0.38]. In two provinces, observed H(3)=8.77,H(3)=8.07,p<0.05], yet posthoc tests revealed pairwise comparisons. Nevertheless, BNs significantly outperformed most training datasets. The results put forward have equal power overall, are best for data-fitting applications.

Language: Английский

---