Research & Publications

Published work in machine learning, computer security, and applied statistics

Research Profile

My research interests span probabilistic machine learning, statistical modelling, spatial analysis, health and wealth inequality, and applied data science.

A recurring theme across my work is trustworthy quantitative decision-making: not only whether a model produces a prediction, but how reliable the result is, how it should be interpreted, and what the analysis tells us about the underlying problem.

Peer-Reviewed Publications

Coronavirus Lung Image Classification with Uncertainty Estimation Using Bayesian Convolutional Neural Networks

Mfundo Monchwe, Ibidun C. Obagbuwa, Alfred Mwanza (2023)

Published in Mathematical Modeling and Intelligent Control for Combating Pandemics, Springer Optimization and Its Applications, Vol. 203, pp. 129–153.

DOI: 10.1007/978-3-031-33183-1_8

Research problem

Many image-classification systems produce probability scores without explicitly modelling uncertainty. In health-related applications, this can create a false sense of confidence around difficult or ambiguous cases.

Contribution

This work investigated Bayesian Convolutional Neural Networks (BCNNs) for coronavirus lung-image classification. The probabilistic approach was used to quantify predictive uncertainty and identify cases where the model may require further inspection.

Technical themes

  • Bayesian deep learning
  • Convolutional neural networks
  • Computer vision
  • Uncertainty estimation
  • Probabilistic machine learning
  • Medical image classification

View publication

Showcasing White-Box Implementation of the RSA Digital Signature Scheme

Colin Chibaya, Mfundo Monchwe, Taryn Nicole Michael, Eli Bila Nimy (2022)

Published in American Journal of Computer Science and Technology, 5(4), 198–203.

DOI: 10.11648/j.ajcst.20220504.11

Research problem

Digital signatures protect the integrity and authenticity of information, but the underlying RSA signing and verification process can be difficult to understand when treated only as a black-box cryptographic procedure.

Contribution

The study presented a white-box implementation of the RSA digital signature scheme and evaluated its behaviour across different parameter and data-length configurations, including the trade-off between security and execution time.

Technical themes

  • RSA
  • Digital signatures
  • Information security
  • Public/private key cryptography
  • Algorithm implementation
  • Performance evaluation

View publication

MSc Research

Spatial and Joint Modelling of Chronic Illness and Household Wealth among South African Adults

MSc Statistics, University of KwaZulu-Natal | Cum Laude

My Master’s research investigated the relationship between chronic illness and household wealth in South Africa while accounting for spatial and nonlinear structure.

The work combined advanced statistical modelling approaches including:

  • Generalized Additive Models
  • Spatial smooths
  • Markov Random Fields
  • Joint modelling
  • Copula-based frameworks
  • Socioeconomic and health-data analysis

The research strengthened my interest in models that move beyond a single global coefficient and instead capture nonlinear relationships, geographic variation, and dependence between outcomes.

Current Research Direction

I continue to develop research around:

  • geoadditive modelling
  • chronic illness and health inequality
  • household wealth
  • joint health-wealth modelling
  • spatial statistics
  • applied statistical learning

Additional manuscripts and research outputs are in development as part of my broader research pipeline.

Research Meets Practice

My research background directly influences how I approach industry data science:

  • I think carefully about data-generating processes, not only algorithms.
  • I distinguish predictive performance from statistical and operational validity.
  • I value uncertainty, interpretability, and reproducibility.
  • I prefer evaluation strategies that reflect how a model will behave on genuinely unseen data.
  • I communicate modelling decisions in terms that technical and non-technical audiences can use.

Profiles

LinkedIn GitHub

Back to top