Artificial intelligence (AI) has dominated the fourth industrial revolution (4IR), and it is the framework for driving many of the automation processes at the centre of the revolution. Statistical learning and machine learning are methodologies used in the AI framework, and aimed at estimating and predicting unknown scientific and business phenomena using pertinent information in data. These methods originated as data set sizes increased, data sources diversified, information dynamics changed and computation power increased.

Statistics is a fundamental contributor to this broad field, and includes research areas such as unstructured and latent data analysis, computer intensive non-parametric methods, and in time predictive modelling. Rapid developments in data analytics necessitates in-depth skills training and research development, balancing fundamental understanding, and research contributions with practical implementation for business and scientific decision-making.

BIOSTATISTICS I and II

The research group strives to find analytical solutions to locally (South African) based problems using innovative methods, or transferring methods developed in other domains to the area of research.

Examples of such solutions included using resampling methods to analyse potable water service operations and port operational efficiencies. In the renewable energy domain, Bayesian tolerance intervals were applied to forecast energy yields in hard to reach areas. The importance for these applications is clear: the results provide information to policy makers who would otherwise use subjective opinions when making decisions. Statistical methods for the analysis of clinical trial data are also important, to ensure that significant differences are based on sound statistical methods.

The UFS Department of Mathematical Statistics and Actuarial Science has built up significant expertise in extreme value analysis and its applications. The main aim of this initiative is to create collaborative inter-institutional network in an effort to increase publication output and build capacity in the field and skills transfer to prospective researchers. This will form a platform for fostering postgraduate students in the area of Extreme Value Theory.

The field of directional statistics received a lot of attention over the past two decades, due to new demands from disciplines such as life sciences and machine learning, the availability of massive data sets requiring adapted statistical techniques, and technological advances. A specific trigger for this research is the need to produce more reliable weather models. Flexible distributions provide powerful insights in the underlying structure of data.

Only a few distributions exist for modelling data on the hyper-sphere, disk or hyper-disc, so there is a need to address this shortcoming. The limitation of current distributions lies in the case where the data appears to not be centred around a specific modal region, but rather possess bimodal behaviour and includes both a mode and an anti-mode. To date, no papers have appeared in which distributions flexible enough to handle bimodal behaviour were proposed for the hyper-disc and other related circular manifolds.