Comparing Bayesian statistical modelling with machine learning in spatio-temporal disease mapping

Abstract

I have decided to embark on a new project to deepen my knowledge of Bayesian inference and space-time modelling. I am particularly interested in exploring the use of the Integrated Nested Laplace Approximation (INLA) methodology, which allows for fast and accurate approximations of posterior distributions, making it an ideal tool for analyzing large and complex datasets. Additionally, I plan to compare classical machine learning models such as Extreme Gradient Boosting or Random Forest and deep learning models such as Long-Short Term Memory (LSTM) or Bayesian Neural Network (BNN) with Bayesian statistical models fitted with INLA to determine their strengths and weaknesses. By identifying which modelling approach is best suited for different types of datasets and analysis tasks, I aim to become a more versatile data analyst. To these ends, we first introduce the theoretical framework explaining the concepts of Bayesian inference, classical machine learning and deep learning in Chapter 2. In Chapter 3, we perform an exploratory data analysis to gain a better understanding of the problem we are facing. Subsequently, rate modelling is presented in Chapter 4, where we outline the advantages and drawbacks of each method. We end this work in Chapter 5 with the conclusions and ideas on further work.