Stochastic filtering based transmissibility estimation of novel coronavirus

Bansal, Rakesh Kumar; Kumar, Amit; Singh, Amit Kumar; Kumar, Sandeep

doi:10.1016/j.dsp.2021.103001

Cited by 7 publications

(2 citation statements)

References 47 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…India was excluded owing to the high prevalence of the new B. 1.617 variant which has increased transmissibility 24 . Although replacing India with Russia as an additional training data set was considered the lack of data available made this unfeasible.…”

Section: Methodsmentioning

confidence: 99%

Machine Learning Model for Predicting Number of COVID19 Cases in Countries with Low Number of Tests

Hashim

Farooq

Syriopoulos

et al. 2021

Preprint

View full text Add to dashboard Cite

The COVID-19 pandemic has presented a series of new challenges to governments and health care systems. Testing is one important method for monitoring and therefore controlling the spread of COVID-19. Yet with a serious discrepancy in the resources available between rich and poor countries not every country is able to employ widespread testing. Here we developed machine learning models for predicting the number of COVID-19 cases in a country based on multilinear regression and neural networks models. The models are trained on data from US states and tested against the reported infections in the European countries. The model is based on four features: Number of tests Population Percentage Urban Population and Gini index. The population and number of tests have the strongest correlation with the number of infections. The model was then tested on data from European countries for which the correlation coefficient between the actual and predicted cases R2 was found to be 0.88 in the multi linear regression and 0.91 for the neural network model. The model predicts that the actual number of infections in countries where the number of tests is less than 10% of their populations is at least 26 times greater than the reported numbers.

show abstract

Section: Methodsmentioning

confidence: 99%

Machine Learning Model for Predicting Number of COVID19 Cases in Countries with Low Number of Tests

Hashim

Farooq

Syriopoulos

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Tracking the reproduction number (

) with confidence bounds based on the KF has been introduced in [ 12 ]. In addition, some literature has compared the performance of COVID-19 prediction with different types of KFs, e.g., fractional-order EKF with an SEIR model [ 13 ], switching KF with time-series models [ 14 ], cubature KF with the SEIRRPV model (Susceptible–Exposed–Infected–Recovered from exposure–Recovered from infection–Passed away–Vaccinated) in [ 15 ] and Quadratic KF with SEIR/ARIMA (AutoRegressive Integrated Moving Average) models [ 16 ]. The novelty of this work as opposed to the existing Kalman filtering works on COVID-19 data is that it reports a thorough benchmarking of the state estimation performances using four alternative formations of the EKF with and without correlated noise and skewed distribution based on the residual of the deterministic model and real data.…”

Section: Introductionmentioning

confidence: 99%

Bayesian Noise Modelling for State Estimation of the Spread of COVID-19 in Saudi Arabia with Extended Kalman Filters

Alyami

Panda

Das

2023

Sensors

View full text Add to dashboard Cite

The epistemic uncertainty in coronavirus disease (COVID-19) model-based predictions using complex noisy data greatly affects the accuracy of pandemic trend and state estimations. Quantifying the uncertainty of COVID-19 trends caused by different unobserved hidden variables is needed to evaluate the accuracy of the predictions for complex compartmental epidemiological models. A new approach for estimating the measurement noise covariance from real COVID-19 pandemic data has been presented based on the marginal likelihood (Bayesian evidence) for Bayesian model selection of the stochastic part of the Extended Kalman filter (EKF), with a sixth-order nonlinear epidemic model, known as the SEIQRD (Susceptible–Exposed–Infected–Quarantined–Recovered–Dead) compartmental model. This study presents a method for testing the noise covariance in cases of dependence or independence between the infected and death errors, to better understand their impact on the predictive accuracy and reliability of EKF statistical models. The proposed approach is able to reduce the error in the quantity of interest compared to the arbitrarily chosen values in the EKF estimation.

show abstract

Unknown uncertainties in the COVID-19 pandemic: Multi-dimensional identification and mathematical modelling for the analysis and estimation of the casualties

Tutsoy

Balıkçı

Ozdil

2021

Digital Signal Processing

View full text Add to dashboard Cite

Insights about the dominant dynamics, coupled structures and the unknown uncertainties of the pandemic diseases play an important role in determining the future characteristics of the pandemic diseases. To enhance the prediction capabilities of the models, properties of the unknown uncertainties in the pandemic disease, which can be utterly random, or function of the system dynamics, or it can be correlated with an unknown function, should be determined. The known structures and amount of the uncertainties can also help the state authorities to improve the policies based on the recognized source of the uncertainties. For instance, the uncertainties correlated with an unknown function imply existence of an undetected factor in the casualties. In this paper, we extend the SpID-N (Suspicious-Infected-Death with non-pharmacological policies) model as in the form of MIMO (Multi-Input-Multi-Output) structure by adding the multi-dimensional unknown uncertainties. The results confirm that the infected and death sub-models mostly have random uncertainties (due undetected casualties) whereas the suspicious sub-model has uncertainties correlated with the internal dynamics (governmental policy of increasing the number of the daily tests) for Turkey. However, since the developed MIMO model parameters are learned from the data (daily reported casualties), it can be easily adapted for other countries. Obtained model with the corresponding uncertainties predicts a distinctive second peak where the number of deaths, infected and suspicious casualties disappear in 240, 290, and more than 300 days, respectively, for Turkey.

show abstract

Stochastic filtering based transmissibility estimation of novel coronavirus

Cited by 7 publications

References 47 publications

Machine Learning Model for Predicting Number of COVID19 Cases in Countries with Low Number of Tests

Machine Learning Model for Predicting Number of COVID19 Cases in Countries with Low Number of Tests

Bayesian Noise Modelling for State Estimation of the Spread of COVID-19 in Saudi Arabia with Extended Kalman Filters

Unknown uncertainties in the COVID-19 pandemic: Multi-dimensional identification and mathematical modelling for the analysis and estimation of the casualties

Contact Info

Product

Resources

About