A Flexible Data-Driven Framework for COVID-19 Case Forecasting Deployed in a Developing-world Public Health Setting

Forecasting infection case counts and estimating accurate epidemiological parameters are critical components of managing the response to a pandemic. This paper describes a modular, extensible framework for a COVID-19 forecasting system, primarily deployed in Mumbai and Jharkhand, India. We employ a variant of the SEIR compartmental model motivated by the nature of the available data and operational constraints. We estimate best-fit parameters using sequential Model-Based Optimization (SMBO) and describe the use of a novel, fast, and approximate Bayesian model averaging method (ABMA) for parameter uncertainty estimation that compares well with a more rigorous Markov Chain Monte Carlo (MCMC) approach in practice. We address on-the-ground deployment challenges such as spikes in the reported input data using a novel weighted smooth-ing method. We describe extensive empirical analyses to evaluate the accuracy of our method on ground truth as well as against other state-of-the-art approaches. Finally, we outline deployment lessons and describe how inferred model parameters were used by government partners to interpret the state of the epidemic and how model forecasts were used to estimate staffing and planning needs essential for addressing COVID-19 hospital burden.

Download Full-text

Distinguishing Normative Processes From Noise

Social Psychology Quarterly ◽

10.1177/0190272516666794 ◽

2016 ◽

Vol 79 (4) ◽

pp. 311-332 ◽

Cited By ~ 4

Author(s):

Jonathan H. Morgan ◽

Kimberly B. Rogers ◽

Mao Hu

Keyword(s):

Bayesian Methods ◽

Bayesian Model ◽

Stepwise Regression ◽

Bayesian Model Averaging ◽

Model Averaging ◽

Ground Truth ◽

Model Specification ◽

Affect Control Theory ◽

False Discovery ◽

Social Events

This research evaluates the relative merits of two established and two newly proposed methods for modeling impressions of social events: stepwise regression, ANOVA, Bayesian model averaging, and Bayesian model sampling. Models generated with each method are compared against a ground truth model to assess performance at variable selection and coefficient estimation. We also assess the theoretical impacts of different modeling choices. Results show that the ANOVA procedure has a significantly lower false discovery rate than stepwise regression, whereas Bayesian methods exhibit higher true positive rates and comparable false discovery rates to ANOVA. Bayesian methods also generate coefficient estimates with less bias and variance than either stepwise regression or ANOVA. We recommend the use of Bayesian methods for model specification in affect control theory.

Download Full-text

Bayesian Solutions for Handling Uncertainty in Survival Extrapolation

Medical Decision Making ◽

10.1177/0272989x16650669 ◽

2016 ◽

Vol 37 (4) ◽

pp. 367-376 ◽

Cited By ~ 8

Author(s):

Miguel A. Negrín ◽

Julian Nam ◽

Andrew H. Briggs

Keyword(s):

Model Uncertainty ◽

Bayesian Model Averaging ◽

Decision Time ◽

Information Criteria ◽

Model Parameters ◽

Posterior Probabilities ◽

Underlying Distribution ◽

Considerable Uncertainty ◽

Best Fit

Objective. Survival extrapolation using a single, best-fit model ignores 2 sources of model uncertainty: uncertainty in the true underlying distribution and uncertainty about the stability of the model parameters over time. Bayesian model averaging (BMA) has been used to account for the former, but it can also account for the latter. We investigated BMA using a published comparison of the Charnley and Spectron hip prostheses using the original 8-year follow-up registry data. Methods. A wide variety of alternative distributions were fitted. Two additional distributions were used to address uncertainty about parameter stability: optimistic and skeptical. The optimistic (skeptical) model represented the model distribution with the highest (lowest) estimated probabilities of survival but reestimated using, as prior information, the most optimistic (skeptical) parameter estimated for intermediate follow-up periods. Distributions were then averaged assuming the same posterior probabilities for the optimistic, skeptical, and noninformative models. Cost-effectiveness was compared using both the original 8-year and extended 16-year follow-up data. Results. We found that all models obtained similar revision-free years during the observed period. In contrast, there was variability over the decision time horizon. Over the observed period, we detected considerable uncertainty in the shape parameter for Spectron. After BMA, Spectron was cost-effective at a threshold of £20,000 with 93% probability, whereas the best-fit model was 100%; by contrast, with a 16-year follow-up, it was 0%. Conclusions. This case study casts doubt on the ability of the single best-fit model selected by information criteria statistics to adequately capture model uncertainty. Under this scenario, BMA weighted by posterior probabilities better addressed model uncertainty. However, there is still value in regularly updating health economic models, even where decision uncertainty is low.

Download Full-text

Bayesian Model Averaging and Prior Sensitivity in Stochastic Frontier Analysis

Econometrics ◽

10.3390/econometrics8020013 ◽

2020 ◽

Vol 8 (2) ◽

pp. 13 ◽

Cited By ~ 1

Author(s):

Kamil Makieła ◽

Błażej Mazur

Keyword(s):

Latent Variables ◽

Bayesian Model ◽

Stochastic Frontier Analysis ◽

Bayesian Model Averaging ◽

Scale Parameter ◽

Stochastic Frontier ◽

Model Averaging ◽

Model Parameters ◽

Frontier Analysis ◽

Stochastic Parameters

This paper discusses Bayesian model averaging (BMA) in Stochastic Frontier Analysis and investigates inference sensitivity to prior assumptions made about the scale parameter of (in)efficiency. We turn our attention to the “standard” prior specifications for the popular normal-half-normal and normal-exponential models. To facilitate formal model comparison, we propose a model that nests both sampling models and generalizes the symmetric term of the compound error. Within this setup it is possible to develop coherent priors for model parameters in an explicit way. We analyze sensitivity of different prior specifications on the aforementioned scale parameter with respect to posterior characteristics of technology, stochastic parameters, latent variables and—especially—the models’ posterior probabilities, which are crucial for adequate inference pooling. We find that using incoherent priors on the scale parameter of inefficiency has (i) virtually no impact on the technology parameters; (ii) some impact on inference about the stochastic parameters and latent variables and (iii) substantial impact on marginal data densities, which are crucial in BMA.

Download Full-text

Bayesian System Averaging: A Grand Theory Unifying Bayesian Forecasting System Method and Bayesian Model Averaging Method

Systems Engineering in Context ◽

10.1007/978-3-030-00114-8_54 ◽

2019 ◽

pp. 677-690

Author(s):

Jie Liu

Keyword(s):

Bayesian Model ◽

Bayesian Model Averaging ◽

Averaging Method ◽

Model Averaging ◽

Bayesian Forecasting ◽

Grand Theory ◽

System Method ◽

Forecasting System

Download Full-text

Cost-Effective Prediction in Medicine and Marketing: Only the Difference Between Bayesian Model Averaging and the Single Best-Fit Model

2019 IEEE 31st International Conference on Tools with Artificial Intelligence (ICTAI) ◽

10.1109/ictai.2019.00178 ◽

2019 ◽

Author(s):

Paul Darwen

Keyword(s):

Bayesian Model ◽

Bayesian Model Averaging ◽

Model Averaging ◽

Cost Effective ◽

The Difference ◽

Best Fit

Download Full-text

Computing and Using Inclusion Bayes Factors for Mixed Fixed and Random Effect Diffusion Decision Models

10.31234/osf.io/45t2w ◽

2021 ◽

Author(s):

Udo Boehm ◽

Nathan J. Evans ◽

Quentin Frederik Gronau ◽

Dora Matzke ◽

Eric-Jan Wagenmakers ◽

...

Keyword(s):

Bayesian Model Averaging ◽

Random Effect ◽

Quantitative Description ◽

Model Averaging ◽

Cognitive Models ◽

Parameter Estimates ◽

Model Parameters ◽

Data Sets ◽

Bayesian Hierarchical ◽

Selective Influence

Cognitive models provide a substantively meaningful quantitative description of latent cognitive processes. The quantitative formulation of these models supports cumulative theory building and enables strong empirical tests. However, the non-linearity of these models and pervasive correlations among model parameters pose special challenges when applying cognitive models to data. Firstly, estimating cognitive models typically requires large hierarchical data sets that need to be accommodated by an appropriate statistical structure within the model. Secondly, statistical inference needs to appropriately account for model uncertainty to avoid overconfidence and biased parameter estimates. In the present work we show how these challenges can be addressed through a combination of Bayesian hierarchical modelling and Bayesian model averaging. To illustrate these techniques, we apply the popular diffusion decision model to data from a collaborative selective influence study.

Download Full-text

Factors Affecting Life Expectancy in East Java: Predictions with A Bayesian Model Averaging Approach

Jurnal Perencanaan Pembangunan: The Indonesian Journal of Development Planning ◽

10.36574/jpp.v5i2.214 ◽

2021 ◽

Vol 5 (2) ◽

pp. 283-295

Author(s):

Harun Al Azies ◽

Vivi Mentari Dewi

Keyword(s):

Life Expectancy ◽

Regression Model ◽

Bayesian Model ◽

Bayesian Model Averaging ◽

Model Averaging ◽

Infant Mortality Rate ◽

Model Parameters ◽

Data Set ◽

Expectancy Value ◽

Factors Affecting

This study predicts the factors that influence life expectancy in East Java, Indonesia. In particular, this study compares the prediction results between the linear regression model and the Bayesian Model Averaging (BMA). The study used a 2015 data set from the Central Bureau of Statistics (BPS) of the province of East Java.The results of data exploration show that the life expectancy in East Java is 70.68 years, the Bondowoso regency is the region with the lowest life expectancy at 65.73 years and the city of Surabaya is the area with the highest life expectancy value in East Java, which is 73.85 years.The results of the inference study indicate that the factors that are expected to affect life expectancy in East Java are the infant mortality rate and the illiteracy rate of the population aged 10 and over.The results of the comparison between the BMA and the regression show that the BMA is a better model for predicting the factors that affect life expectancy in East Java than the regression model because the BMA model can estimate the parameters more efficiently by estimating the model parameters based on the standard error value.

Download Full-text

Simulation of Gross Primary Productivity Using Multiple Light Use Efficiency Models

Land ◽

10.3390/land10030329 ◽

2021 ◽

Vol 10 (3) ◽

pp. 329

Author(s):

Jun Zhang ◽

Xufeng Wang ◽

Jun Ren

Keyword(s):

Remote Sensing ◽

Primary Productivity ◽

Bayesian Model Averaging ◽

Model Averaging ◽

Gross Primary Productivity ◽

Accurate Estimation ◽

Model Parameters ◽

Light Use Efficiency ◽

Multiple Model ◽

Use Efficiency

Gross primary productivity (GPP) is the most basic variable in a carbon cycle study that determines the carbon that enters the ecosystem. The remote sensing-based light use efficiency (LUE) model is one of the primary tools that is currently used to estimate the GPP at the regional scale. Many remote sensing-based GPP models have been developed in the last several decades, and these models have been well evaluated at some sites. However, an accurate estimation of the GPP remains challenging work using LUE models because of uncertainties in the model caused by model parameters, model forcing, and vegetation spatial heterogeneity. In this study, five widely used LUE models, Glo-PEM, VPM, EC-LUE, the MODIS GPP algorithm, and C-fix, were selected to simulate the GPP of the Heihe River Basin forced using in situ measurements. A multiple-model averaging method, Bayesian model averaging (BMA), was used to combine the five models to obtain a more reliable GPP estimation. The BMA was trained using carbon flux data from five eddy covariance towers located at dominant vegetation types in the study area. Generally, the BMA method performed better than any single LUE model. From the case study in the study area, it is indicated that the trained BMA is an efficient method to combine multiple LUE models and can improve the GPP simulation accuracy.

Download Full-text