Researcher: Vhahangwele Netshilonwe, Sol Plaatje University
Supervisor: Dr. Michael Olusanya , Sol Plaatje University

This study investigates how machine learning can be used to detect road damage, specifically potholes, in images of roads. The study tests a particular machine learning model called the “one-class support vector
machine (OCSVM)” to determine if it can effectively identify road damage better than random chance. The objective is to enhance our capability to detect and address road damage problems.

Researcher: Mafanedza Nephawe, University of the Witwatersrand, Johannesburg
Supervisor: Prof Prince Ntimeni

The Higgs boson, a fundamental particle, gives mass to others and was confirmed in 2012 at
CERN[1, 5]. The associated Higgs field explains mass differences and supports the Standard
Model. [2]Large Hadron Collider channels target specific particles, ZH (associated production of
top quarks with Higgs and Z bosons). This project introduces CEPC, a circular electron-positron
collider in China, aiming to study particle properties[5, 3]. AI and machine learning analyze detector
data to understand the Higgs boson and explore beyond the standard model. The research
aims to employ a deep neural network for model-independent analysis of the Standard Model
Higgs, focusing on signal-background separation in ZH channels at CEPC with a 40% branching
ratio at 95 GeV.

Researcher: Mafanedza Nephawe, University of the Witwatersrand, Johannesburg
Supervisor: Dr. M. Arasomwan

Researcher: Dzhivhuho Asakundwi Praisethelord, University of Venda
Supervisor: Dr. M. Arasomwan

Researcher: Vincent Mohale, Sol Plaatjie University
Supervisor: Dr Ibidun Obagbuwa

Bloemfontein, a city, where wheat has long been a culinary cornerstone, faces a silent transformation which is underway. Wheat, the bedrock of staples like bread, cereals, and the essence of beverages such as beer and whiskey, now faces a challenging foe: climatechange.
In this study, we delve into the heart of this global staple, unravelling the complex web of connections that link climate change to wheat yield in Bloemfontein.

Researcher: Jaden Pieterse, Sol Plaatjie University
Supervisor: Dr Martins Arosamwan, University of the Witwatersrand, Johannesburg

The need for accurate demand and sales forecasting is important for companies to perform better. Due to rapid technological development, e-commerce platforms face challenges with finding proper models to deal with market demand. LSTM is considered to perform best in predicting demand.

Researcher: Joey J. Assabil
Supervisor: Dr C. Obagbuwa

Credit Card Fraud (CCF) hasbeen a worldwide conundrumresulting in millions of losses.The research compares theefficiency of Traditional algorithms (LogisticRegression, Decision Trees,K-Nearest Neighbor) andEnsemble algorithms(Random Forest, Adaboost,Xgboost) in detectingfraudulent patterns withindatasets.

Researcher: Kgoale Tshiamo Mahlako
Supervisor: Dr A Whata

Survival analysis is a vital field in statistics, offering valuable insights into various domains, from healthcare to epidemiology. It allows us to understand the factors that influence time-to-event outcomes. The analysis of survival data comes with its own set of challenges, especially when it involves censoring.
Censoring occurs when we don’t have complete information about the event times, making it difficult to draw causal inferences. The application of pseudo-observations, which is a crucial tool for enhancing causal inference, will be the subject of our particular attention. We apply the G-formula and IPTW, methods from causal inference, on these pseudo-observations. And estimate the ATE for a completely observed outcome and censored data.

Researcher: Emelia Thembile Kammies
Supervisor: Dr Lyson Chaka

Data mining, a vital machine learning technique, extracts valuable information from raw data across diverse fields like healthcare, retail, logistics, military, banking, and sports. It employs supervised and unsupervised methods, with classification being the most popular. This study evaluated six common machine learning classification models across six different datasets and identified their strengths.