arXiv subjects

Theo Economou

Publications and source records attributed to Theo Economou.

9 recordsLinked to original sources

arXiv2024

A unifying modelling approach for hierarchical distributed lag models

We present a statistical modelling framework for implementing Distributed Lag Models (DLMs), encompassing several extensions of the approach to capture the temporally distributed effect from covariates via regression. We place DLMs in the context of penalised Generalized Additive Models (GAMs) and illustrate that implementation via the R package \texttt{mgcv}, which allows for flexible and interpretable inference in addition to thorough model assessment. We show how the interpretation of penalised splines as random quantities enables approximate Bayesian inference and hierarchical structures in the same practical setting. We focus on epidemiological studies and demonstrate the approach with application to mortality data from Cyprus and Greece. For the Cyprus case study, we investigate for the first time, the joint lagged effects from both temperature and humidity on mortality risk with the unexpected result that humidity severely increases risk during cold rather than hot conditions. Another novel application is the use of the proposed framework for hierarchical pooling, to estimate district-specific covariate-lag risk on morality and the use of posterior simulation to compare risk across districts.

stat.ME↗

arXiv2022

A deep mixture density network for outlier-corrected interpolation of crowd-sourced weather data

As the costs of sensors and associated IT infrastructure decreases - as exemplified by the Internet of Things - increasing volumes of observational data are becoming available for use by environmental scientists. However, as the number of available observation sites increases, so too does the opportunity for data quality issues to emerge, particularly given that many of these sensors do not have the benefit of official maintenance teams. To realise the value of crowd sourced 'Internet of Things' type observations for environmental modelling, we require approaches that can automate the detection of outliers during the data modelling process so that they do not contaminate the true distribution of the phenomena of interest. To this end, here we present a Bayesian deep learning approach for spatio-temporal modelling of environmental variables with automatic outlier detection. Our approach implements a Gaussian-uniform mixture density network whose dual purposes - modelling the phenomenon of interest, and learning to classify and ignore outliers - are achieved simultaneously, each by specifically designed branches of our neural network. For our example application, we use the Met Office's Weather Observation Website data, an archive of observations from around 1900 privately run and unofficial weather stations across the British Isles. Using data on surface air temperature, we demonstrate how our deep mixture model approach enables the modelling of a highly skilled spatio-temporal temperature distribution without contamination from spurious observations. We hope that adoption of our approach will help unlock the potential of incorporating a wider range of observation sources, including from crowd sourcing, into future environmental models.

stat.ML↗

arXiv2020

A Powerful Modelling Framework for Nowcasting and Forecasting COVID-19 and Other Diseases

The COVID-19 pandemic has highlighted delayed reporting as a significant impediment to effective disease surveillance and decision-making. In the absence of timely data, statistical models which account for delays can be adopted to nowcast and forecast cases or deaths. We discuss the four key sources of systematic and random variability in available data for COVID-19 and other diseases, and critically evaluate current state-of-the-art methods with respect to appropriately separating and capturing this variability. We present a general spatio-temporal hierarchical framework for correcting delayed reporting and demonstrate its application to daily English hospital deaths from COVID-19 and Severe Acute Respiratory Infection cases in Brazil. We compare our approach to competing models with respect to theoretical flexibility and quantitative metrics from a rolling nowcasting experiment imitating a realistic operational scenario. Based on consistent and compelling leads in nowcasting accuracy, bias, and precision, we demonstrate that our approach represents the current best-practice for correcting delayed reporting.

stat.AP↗

arXiv2020

Bayesian deep learning for mapping via auxiliary information: a new era for geostatistics?

For geospatial modelling and mapping tasks, variants of kriging - the spatial interpolation technique developed by South African mining engineer Danie Krige - have long been regarded as the established geostatistical methods. However, kriging and its variants (such as regression kriging, in which auxiliary variables or derivatives of these are included as covariates) are relatively restrictive models and lack capabilities that have been afforded to us in the last decade by deep neural networks. Principal among these is feature learning - the ability to learn filters to recognise task-specific patterns in gridded data such as images. Here we demonstrate the power of feature learning in a geostatistical context, by showing how deep neural networks can automatically learn the complex relationships between point-sampled target variables and gridded auxiliary variables (such as those provided by remote sensing), and in doing so produce detailed maps of chosen target variables. At the same time, in order to cater for the needs of decision makers who require well-calibrated probabilities, we obtain uncertainty estimates via a Bayesian approximation known as Monte Carlo dropout. In our example, we produce a national-scale probabilistic geochemical map from point-sampled assay data, with auxiliary information provided by a terrain elevation grid. Unlike traditional geostatistical approaches, auxiliary variable grids are fed into our deep neural network raw. There is no need to provide terrain derivatives (e.g. slope angles, roughness, etc) because the deep neural network is capable of learning these and arbitrarily more complex derivatives as necessary to maximise predictive performance. We hope our results will raise awareness of the suitability of Bayesian deep learning - and its feature learning capabilities - for large-scale geostatistical applications where uncertainty matters.

stat.ML↗

arXiv2020

A framework for probabilistic weather forecast post-processing across models and lead times using machine learning

Forecasting the weather is an increasingly data intensive exercise. Numerical Weather Prediction (NWP) models are becoming more complex, with higher resolutions, and there are increasing numbers of different models in operation. While the forecasting skill of NWP models continues to improve, the number and complexity of these models poses a new challenge for the operational meteorologist: how should the information from all available models, each with their own unique biases and limitations, be combined in order to provide stakeholders with well-calibrated probabilistic forecasts to use in decision making? In this paper, we use a road surface temperature example to demonstrate a three-stage framework that uses machine learning to bridge the gap between sets of separate forecasts from NWP models and the 'ideal' forecast for decision support: probabilities of future weather outcomes. First, we use Quantile Regression Forests to learn the error profile of each numerical model, and use these to apply empirically-derived probability distributions to forecasts. Second, we combine these probabilistic forecasts using quantile averaging. Third, we interpolate between the aggregate quantiles in order to generate a full predictive distribution, which we demonstrate has properties suitable for decision support. Our results suggest that this approach provides an effective and operationally viable framework for the cohesive post-processing of weather forecasts across multiple models and lead times to produce a well-calibrated probabilistic output.

stat.AP↗

arXiv2019

Global Household Energy Model: A Multivariate Hierarchical Approach to Estimating Trends in the Use of Polluting and Clean Fuels for Cooking

In 2017 an estimated 3 billion people used polluting fuels and technologies as their primary cooking solution, with 3.8 million deaths annually attributed to household exposure to the resulting fine particulate matter air pollution. Currently, health burdens are calculated using aggregations of fuel types, e.g. solid fuels, as country-level estimates of the use of specific fuel types, e.g. wood and charcoal, are unavailable. To expand the knowledge base about impacts of household air pollution on health, we develop and implement a Bayesian hierarchical model, based on Generalized Dirichlet Multinomial distributions, that jointly estimates non-linear trends in the use of eight key fuel types, overcoming several data-specific challenges including missing or combined fuel use values. We assess model fit using within-sample predictive analysis and an out-of-sample prediction experiment to evaluate the model's forecasting performance.

stat.AP↗

arXiv2019

An Advanced Hidden Markov Model for Hourly Rainfall Time Series

For hydrological applications, such as urban flood modelling, it is often important to be able to simulate sub-daily rainfall time series from stochastic models. However, modelling rainfall at this resolution poses several challenges, including a complex temporal structure including long dry periods, seasonal variation in both the occurrence and intensity of rainfall, and extreme values. We illustrate how the hidden Markov framework can be adapted to construct a compelling model for sub-daily rainfall, which is capable of capturing all of these important characteristics well. These adaptations include clone states and non-stationarity in both the transition matrix and conditional models. Set in the Bayesian framework, a rich quantification of both parametric and predictive uncertainty is available, and thorough model checking is made possible through posterior predictive analyses. Results from the model are interpretable, allowing for meaningful examination of seasonal variation and medium to long term trends in rainfall occurrence and intensity. To demonstrate the effectiveness of our approach, both in terms of model fit and interpretability, we apply the model to an 8-year long time series of hourly observations.

stat.AP↗

arXiv2019

Multivariate Hierarchical Frameworks for Modelling Delayed Reporting in Count Data

In many fields and applications count data can be subject to delayed reporting. This is where the total count, such as the number of disease cases contracted in a given week, may not be immediately available, instead arriving in parts over time. For short term decision making, the statistical challenge lies in predicting the total count based on any observed partial counts, along with a robust quantification of uncertainty. In this article we discuss previous approaches to modelling delayed reporting and present a multivariate hierarchical framework where the count generating process and delay mechanism are modelled simultaneously. Unlike other approaches, the framework can also be easily adapted to allow for the presence of under-reporting in the final observed count. To compare our approach with existing frameworks, one of which we extend to potentially improve predictive performance, we present a case study of reported dengue fever cases in Rio de Janeiro. Based on both within-sample and out-of-sample posterior predictive model checking and arguments of interpretability, adaptability, and computational efficiency, we discuss the advantages and disadvantages of each modelling framework.

stat.AP↗

arXiv2018

A Hierarchical Framework for Correcting Under-Reporting in Count Data

Tuberculosis poses a global health risk and Brazil is among the top twenty countries by absolute mortality. However, this epidemiological burden is masked by under-reporting, which impairs planning for effective intervention. We present a comprehensive investigation and application of a Bayesian hierarchical approach to modelling and correcting under-reporting in tuberculosis counts, a general problem arising in observational count data. The framework is applicable to fully under-reported data, relying only on an informative prior distribution for the mean reporting rate to supplement the partial information in the data. Covariates are used to inform both the true count generating process and the under-reporting mechanism, while also allowing for complex spatio-temporal structures. We present several sensitivity analyses based on simulation experiments to aid the elicitation of the prior distribution for the mean reporting rate and decisions relating to the inclusion of covariates. Both prior and posterior predictive model checking are presented, as well as a critical evaluation of the approach.

stat.AP↗