Projects

The Covid-19 pandemic has clearly shown the use of data and extraction of information from such data for making decisions. This project will consider some part of all the data available and use statistical methods for predicting number of hospital incidences from test data.

Within this project applied data analysis and predictive modeling will be carried out. A student is allowed to choose a competition or a data-set of interest for him/her on one of the popular data science platforms*: kaggle, topcoder or uci. Then preliminary data analysis should be performed, followed by careful statistical modeling, inference and eventually evaluation of predictions and explaining the results. The final report should be delivered in latex and should include description of the data and problem, the choice and specification of an appropriate statistical model, evaluation of the model in terms of predictions and (if applicable) explanations of the model.

Fractional polynomials is a popular class of statistical models that creates an alternative to regular polynomials and allow flexible parameterization of continuous variables. There exist several implementation of fractional polynomials in the literature. The aim of this project would be to prepare and carry out a small simulation study, where the ability of the approaches to recover the correct polynomials will be studied. Also the student can perform comparison of predictive performance of the studied approaches on a real data set.

Integration occurs as a numerical problem in many settings. For high-dimensional problems, Monte Carlo  is a useful class of methods for performing such  problems. In this project the aim is to find relevant literature about what Monte Carlo is, how it is used and their properties.

Michael Baumgartner and Christoph Falk from the University of Bergen recently published a paper entitled Configurational causal modeling and logic regression [1]. In this paper, they show that logic regression model can be used to recover causal structures featuring conjunctural causation and equifinality. Further they hypothesise that Bayesian logic regression may also give promising results. The aim of this project, thus, would be to check if a Bayesian version of logic regression can perform the task better than the frequentist version that is used by Michael Baumgartner and Christoph Falk. The project will involve carefully reading through the paper by Baumgartner and Falk, understanding the problem and repeating their simulation study (their code and data are available) with a version of Bayesian logic regression implemented in EMJMCMC R package or in LogicReg package.