arXiv subjects

Kevin Burke

Publications and source records attributed to Kevin Burke.

At least 19 recordsLinked to original sources

arXiv2025

Comparison of generalised additive models and neural networks in applications: A systematic review

Neural networks have become a popular tool in predictive modelling, more commonly associated with machine learning and artificial intelligence than with statistics. Generalised Additive Models (GAMs) are flexible non-linear statistical models that retain interpretability. Both are state-of-the-art in their own right, with their respective advantages and disadvantages. This paper analyses how these two model classes have performed on real-world tabular data. Following PRISMA guidelines, we conducted a systematic review of papers that performed empirical comparisons of GAMs and neural networks. Eligible papers were identified, yielding 143 papers, with 430 datasets. Key attributes at both paper and dataset levels were extracted and reported. Beyond summarising comparisons, we analyse reported performance metrics using mixed-effects modelling to investigate potential characteristics that can explain and quantify observed differences, including application area, study year, sample size, number of predictors, and neural network complexity. Across datasets, no consistent evidence of superiority was found for either GAMs or neural networks when considering the most frequently reported metrics (RMSE, $R^2$, and AUC). Neural networks tended to outperform in larger datasets and in those with more predictors, but this advantage narrowed over time. Conversely, GAMs remained competitive, particularly in smaller data settings, while retaining interpretability. Reporting of dataset characteristics and neural network complexity was incomplete in much of the literature, limiting transparency and reproducibility. This review highlights that GAMs and neural networks should be viewed as complementary approaches rather than competitors. For many tabular applications, the performance trade-off is modest, and interpretability may favour GAMs.

stat.ML↗

arXiv2025

A General Simulation-Based Optimisation Framework for Multipoint Constant-Stress Accelerated Life Tests

Accelerated life testing (ALT) is a method of reducing the lifetime of components through exposure to extreme stress. This method of obtaining lifetime information involves the design of a testing experiment, i.e., an accelerated test plan. In this work, we adopt a simulation-based approach to obtaining optimal test plans for constant-stress accelerated life tests with multiple design points. Within this simulation framework we can easily assess a variety of test plans by modifying the number of test stresses (and their levels) and evaluating the allocation of test units. We obtain optimal test plans by utilising the differential evolution (DE) optimisation algorithm, where the inputs to the objective function are the test plan parameters, and the output is the RMSE (root mean squared error) of out-of-sample (extrapolated) model predictions. When the life-stress distribution is correctly specified, we show that the optimal number of stress levels is related to the number of model parameters. In terms of test unit allocation, we show that the proportion of test units is inversely related to the stress level. Our general simulation framework provides an alternative approach to theoretical optimisation, and is particularly favourable for large/complex multipoint test plans where analytical optimisation could prove intractable. Our procedure can be applied to a broad range of experimental scenarios, and serves as a useful tool to aid practitioners seeking to maximise component lifetime information through accelerated life testing.

stat.AP↗

arXiv2023

Investigating Statistical Inference and Covariate Effects in Shallow Neural Networks

Feedforward neural networks (FNNs) are typically viewed as pure prediction algorithms, and their strong predictive performance has led to their use in many machine-learning applications. However, their flexibility comes with an interpretability trade-off; thus, FNNs have been historically less popular among statisticians. Nevertheless, for suitably parsimonious shallow FNNs, classical statistical theory, such as significance testing and uncertainty quantification, may still provide useful regression-style summaries. Supplementing FNNs with methods of statistical inference, and covariate-effect visualisations, can shift the focus away from black-box prediction and move FNNs towards traditional statistical models. This can allow for more inferential analysis, and, hence, make FNNs more accessible within the statistical-modelling context. We investigate covariate-level Wald testing in the context of penalised FNNs, and also propose covariate-effect plots that emulate regression coefficients. Simulation studies are used to extensively investigate the performance of Wald-based inference, with particular emphasis on when this approach performs well. This statistical-based approach to neural networks is demonstrated through an application to insurance data.

stat.ME↗

arXiv2022

Robust Distributional Regression with Automatic Variable Selection

Datasets with extreme observations and/or heavy-tailed error distributions are commonly encountered and should be analyzed with careful consideration of these features from a statistical perspective. Small deviations from an assumed model, such as the presence of outliers, can cause classical regression procedures to break down, potentially leading to unreliable inferences. Other distributional deviations, such as heteroscedasticity, can be handled by going beyond the mean and modelling the scale parameter in terms of covariates. We propose a method that accounts for heavy tails and heteroscedasticity through the use of a generalized normal distribution (GND). The GND contains a kurtosis-characterizing shape parameter that moves the model smoothly between the normal distribution and the heavier-tailed Laplace distribution - thus covering both classical and robust regression. A key component of statistical inference is determining the set of covariates that influence the response variable. While correctly accounting for kurtosis and heteroscedasticity is crucial to this endeavour, a procedure for variable selection is still required. For this purpose, we use a novel penalized estimation procedure that avoids the typical computationally demanding grid search for tuning parameters. This is particularly valuable in the distributional regression setting where the location and scale parameters depend on covariates, since the standard approach would have multiple tuning parameters (one for each distributional parameter). We achieve this by using a "smooth information criterion" that can be optimized directly, where the tuning parameters are fixed at log(n) in the BIC case.

stat.ME↗

arXiv2022

Analysis of mean-field approximation for Deffuant opinion dynamics on networks

Mean-field equations have been developed recently to approximate the dynamics of the Deffuant model of opinion formation. These equations can describe both fully-mixed populations and the case where individuals interact only along edges of a network. In each case, interactions only occur between individuals whose opinions differ by less than a given parameter, called the confidence bound. The size of the confidence bound parameter is known to strongly affect both the dynamics and the number and location of opinion clusters. In this work we carry out a mathematical analysis of the mean-field equations to investigate the role of the confidence bound and boundaries on these important observables of the model. We consider the limit in which the confidence bound interval is small, and identify the key mechanisms driving opinion evolution. We show that linear stability analysis can predict the number and location of opinion clusters. Comparison with numerical simulations of the model illustrates that the early-time dynamics and the final cluster locations can be accurately approximated for networks composed of two degree classes, as well as for the case of a fully-mixed population.

physics.soc-ph↗

arXiv2022

A Statistical-Modelling Approach to Feedforward Neural Network Model Selection

Feedforward neural networks (FNNs) can be viewed as non-linear regression models, where covariates enter the model through a combination of weighted summations and non-linear functions. Although these models have some similarities to the approaches used within statistical modelling, the majority of neural network research has been conducted outside of the field of statistics. This has resulted in a lack of statistically-based methodology, and, in particular, there has been little emphasis on model parsimony. Determining the input layer structure is analogous to variable selection, while the structure for the hidden layer relates to model complexity. In practice, neural network model selection is often carried out by comparing models using out-of-sample performance. However, in contrast, the construction of an associated likelihood function opens the door to information-criteria-based variable and architecture selection. A novel model selection method, which performs both input- and hidden-node selection, is proposed using the Bayesian information criterion (BIC) for FNNs. The choice of BIC over out-of-sample performance as the model selection objective function leads to an increased probability of recovering the true model, while parsimoniously achieving favourable out-of-sample performance. Simulation studies are used to evaluate and justify the proposed method, and applications on real data are investigated.

stat.ME↗

arXiv2022

Process Visualization of Manufacturing Execution System (MES) Data

Process visualizations of data from manufacturing execution systems (MESs) provide the ability to generate valuable insights for improved decision-making. Industry 4.0 is awakening a digital transformation where advanced analytics and visualizations are critical. Exploiting MESs with data-driven strategies can have a major impact on business outcomes. The advantages of employing process visualizations are demonstrated through an application to real-world data. Visualizations, such as dashboards, enable the user to examine the performance of a production line at a high level. Furthermore, the addition of interactivity facilitates the user to customize the data they want to observe. Evidence of process variability between shifts and days of the week can be investigated with the goal of optimizing production.

cs.HC↗

arXiv2021

Multi-Parameter Regression Survival Modelling with Random Effects

We consider a parametric modelling approach for survival data where covariates are allowed to enter the model through multiple distributional parameters, i.e., scale and shape. This is in contrast with the standard convention of having a single covariate-dependent parameter, typically the scale. Taking what is referred to as a multi-parameter regression (MPR) approach to modelling has been shown to produce flexible and robust models with relatively low model complexity cost. However, it is very common to have clustered data arising from survival analysis studies, and this is something that is under developed in the MPR context. The purpose of this article is to extend MPR models to handle multivariate survival data by introducing random effects in both the scale and the shape regression components. We consider a variety of possible dependence structures for these random effects (independent, shared, and correlated), and estimation proceeds using a h-likelihood approach. The performance of our estimation procedure is investigated by a way of an extensive simulation study, and the merits of our modelling approach are illustrated through applications to two real data examples, a lung cancer dataset and a bladder cancer dataset.

stat.ME↗

arXiv2021

Variable Selection Using a Smooth Information Criterion for Distributional Regression Models

Modern variable selection procedures make use of penalization methods to execute simultaneous model selection and estimation. A popular method is the LASSO (least absolute shrinkage and selection operator), the use of which requires selecting the value of a tuning parameter. This parameter is typically tuned by minimizing the cross-validation error or Bayesian information criterion (BIC) but this can be computationally intensive as it involves fitting an array of different models and selecting the best one. In contrast with this standard approach, we have developed a procedure based on the so-called "smooth IC" (SIC) in which the tuning parameter is automatically selected in one step. We also extend this model selection procedure to the distributional regression framework, which is more flexible than classical regression modelling. Distributional regression, also known as multiparameter regression (MPR), introduces flexibility by taking account of the effect of covariates through multiple distributional parameters simultaneously, e.g., mean and variance. These models are useful in the context of normal linear regression when the process under study exhibits heteroscedastic behaviour. Reformulating the distributional regression estimation problem in terms of penalized likelihood enables us to take advantage of the close relationship between model selection criteria and penalization. Utilizing the SIC is computationally advantageous, as it obviates the issue of having to choose multiple tuning parameters.

stat.ME↗

arXiv2021

Binomial confidence intervals for rare events: importance of defining margin of error relative to magnitude of proportion

Confidence interval performance is typically assessed in terms of two criteria: coverage probability and interval width (or margin of error). In this paper, we assess the performance of four common proportion interval estimators: the Wald, Clopper-Pearson (exact), Wilson and Agresti-Coull, in the context of rare-event probabilities. We define the interval precision in terms of a relative margin of error which ensures consistency with the magnitude of the proportion. Thus, confidence interval estimators are assessed in terms of achieving a desired coverage probability whilst simultaneously satisfying the specified relative margin of error. We illustrate the importance of considering both coverage probability and relative margin of error when estimating rare-event proportions, and show that within this framework, all four interval estimators perform somewhat similarly for a given sample size and confidence level. We identify relative margin of error values that result in satisfactory coverage whilst being conservative in terms of sample size requirements, and hence suggest a range of values that can be adopted in practice. The proposed relative margin of error scheme is evaluated analytically, by simulation, and by application to a number of recent studies from the literature.

stat.AP↗

arXiv2021

An age-structured SEIR model for COVID--19 incidence in Dublin, Ireland with framework for evaluating health intervention cost

Strategies adopted globally to mitigate the threat of COVID-19 have primarily involved lockdown measures with substantial economic and social costs with varying degrees of success. Morbidity patterns of COVID-19 variants have a strong association with age, while restrictive lockdown measures have association with negative mental health outcomes in some age groups. Reduced economic prospects may also afflict some age cohorts more than others. Motivated by this, we propose a model to describe COVID-19 community spread incorporating the role of age-specific social interactions. Through a flexible parameterisation of an age-structured deterministic Susceptible Exposed Infectious Removed (SEIR) model, we provide a means for characterising different forms of lockdown which may impact specific age groups differently. Social interactions are represented through age group to age group contact matrices, which can be trained using available data and are thus locally adapted. This framework is easy to interpret and suitable for describing counterfactual scenarios, which could assist policy makers with regard to minimising morbidity balanced with the costs of prospective suppression strategies. Our work originates from an Irish context and we use disease monitoring data from February 29th 2020 to January 31st 2021 gathered by Irish governmental agencies. We demonstrate how Irish lockdown scenarios can be constructed using the proposed model formulation and show results of retrospective fitting to incidence rates and forward planning with relevant ``what if/instead of'' lockdown counterfactuals with uncertainty quantification. Our formulation is agnostic to a specific locale, in that lockdown strategies in other regions can be straightforwardly encoded using this model. The methods we describe are made publicly available online through an accessible and easy to use web interface.

q-bio.PE↗

arXiv2020

A Generalized Framework for Simultaneous Long-Short Feedback Trading

We present a generalization of the Simultaneous Long-Short (SLS) trading strategy described in recent control literature wherein we allow for different parameters across the short and long sides of the controller; we refer to this new strategy as Generalized SLS (GSLS). Furthermore, we investigate the conditions under which positive gain can be assured within the GSLS setup for both deterministic stock price evolution and geometric Brownian motion. In contrast to existing literature in this area (which places little emphasis on the practical application of SLS strategies), we suggest optimization procedures for selecting the control parameters based on historical data, and we extensively test these procedures across a large number of real stock price trajectories (495 in total). We find that the implementation of such optimization procedures greatly improves the performance compared with fixing control parameters, and, indeed, the GSLS strategy outperforms the simpler SLS strategy in general.

q-fin.TR↗

arXiv2020

A Data-Driven Control-Theoretic Paradigm for Pandemic Mitigation with Application to Covid-19

In this paper, we introduce a new control-theoretic paradigm for mitigating the spread of a virus. To this end, our discrete-time controller, aims to reduce the number of new daily deaths, and consequently, the cumulative number of deaths. In contrast to much of the existing literature, we do not rely on a potentially complex virus transmission model whose equations must be customized to the "particulars" of the pandemic at hand. For new viruses such as Covid-19, the epidemiology driving the modelling process may not be well known and model estimation with limited data may be unreliable. With this motivation in mind, the new paradigm described here is data-driven and, to a large extent, we avoid modelling difficulties by concentrating on just two key quantities which are common to pandemics: the doubling time, denoted by $d(k)$ and the peak day denoted by $θ(k)$. Our numerical studies to date suggest that our appealingly simple model can provide a reasonable fit to real data. Given that time is of the essence during the ongoing global health crisis, the intent of this paper is to introduce this new paradigm to control practitioners and describe a number of new research directions suggested by our current results.

q-bio.PE↗

arXiv2020

A generalised mean-field approximation for the Deffuant opinion dynamics model on networks

When the interactions of agents on a network are assumed to follow the Deffuant opinion dynamics model, the outcomes are known to depend on the structure of the underlying network. This behavior cannot be captured by existing mean-field approximations for the Deffuant model. In this paper, a generalised mean-field approximation is derived that accounts for the effects of network topology on Deffuant dynamics through the degree distribution or community structure of the network. The accuracy of the approximation is examined by comparison with large-scale Monte Carlo simulations on both synthetic and real-world networks.

physics.soc-ph↗

arXiv2020

A likelihood-based approach for cure regression models

We propose a new likelihood-based approach for estimation, inference and variable selection for parametric cure regression models in time-to-event analysis under random right-censoring. In this context, it often happens that some subjects are "cured", i.e., they will never experience the event of interest. Then, the sample of censored observations is an unlabeled mixture of cured and "susceptible" subjects. Using inverse probability censoring weighting (IPCW), we propose a likelihood-based estimation procedure for the cure regression model without making assumptions about the distribution of survival times for the susceptible subjects. The IPCW approach does require a preliminary estimate of the censoring distribution, for which general parametric, semi- or non-parametric approaches can be used. The incorporation of a penalty term in our estimation procedure is straightforward; in particular, we propose L1-type penalties for variable selection. Our theoretical results are derived under mild assumptions. Simulation experiments and real data analysis illustrate the effectiveness of the new approach.

stat.ME↗

arXiv2020

A Generalization of the Classical Kelly Betting Formula to the Case of Temporal Correlation

For sequential betting games, Kelly's theory, aimed at maximization of the logarithmic growth of one's account value, involves optimization of the so-called betting fraction $K$. In this Letter, we extend the classical formulation to allow for temporal correlation among bets. To demonstrate the potential of this new paradigm, for simplicity of exposition, we mainly address the case of a coin-flipping game with even-money payoff. To this end, we solve a problem with memory depth $m$. By this, we mean that the outcomes of coin flips are no longer assumed to be i.i.d.random variables. Instead, the probability of heads on flip $k$ depends on previous flips $k-1,k-2,...,k-m$. For the simplest case of $n$ flips, with $m = 1$, we obtain a closed form solution $K_n$ for the optimal betting fraction. This generalizes the classical result for the memoryless case. That is, instead of fraction $K^* = 2p-1$ which pervades the literature for a coin with probability of heads $p\geq 1/2$, our new fraction $K_n$ depends on both $n$ and the parameters associated with the temporal correlation. Generalizations of these results for $m > 1$ and numerical simulations are also included. Finally, we indicate how the theory extends to time-varying feedback and alternative payoff distributions.

math.OC↗

arXiv2020

Agreement Threshold on Axelrod's model of Cultural Dissemination

Shared opinions are an important feature in the formation of social groups. In this paper, we use the Axelrod model of cultural dissemination to represent opinion-based groups. In the Axelrod model, each agent has a set of features which each holds one of a set of nominally related traits. Survey data, for example, has a similar structure, where each participant answers each of a set of items with responses from a fixed list. We present an alternative method of displaying the Axelrod model by representing it as a bipartite graph, i.e., participants and their responses as separate nodes. This allows us to see which feature-trait combinations are selected in the final state. This visualisation is particularly useful when representing survey data as it illustrates the co-evolution of cultures and opinion-based groups in Axelrod's model of cultural diffusion. We also present a modification to the Axelrod model. A standard finding of the Axelrod model with many features is for all agents to fully agree in one cluster. We introduce an agreement threshold and allow nodes to interact only with those neighbours who are within this threshold (i.e., those with similar opinions) rather than those with any opinion. This method reliably yields a large number of clusters for small agreement thresholds and, importantly, does not limit to single cluster when the number of features grows large. This potentially provides a method for modelling opinion-based groups where as opinions are added, the number of clusters increase.

physics.soc-ph↗

arXiv2019

A modelling methodology for social interaction experiments

Analysis of temporal network data arising from online interactive social experiments is not possible with standard statistical methods because the assumptions of these models, such as independence of observations, are not satisfied. In this paper, we outline a modelling methodology for such experiments where, as an example, we analyse data collected using the Virtual Interaction Application (VIAPPL) --- a software platform for conducting experiments that reveal how social norms and identities emerge through social interaction. We apply our model to show that ingroup favouritism and reciprocity are present in the experiments, and to quantify the strengthening of these behaviours over time. Our method enables us to identify participants whose behaviour is markedly different from the norm. We use the method to provide a visualisation of the data that highlights the level of ingroup favouritism, the strong reciprocal relationships, and the different behaviour of participants in the game. While our methodology was developed with VIAPPL in mind, its usage extends to any type of social interaction data.

stat.AP↗