Research
Finding useful answers in complex data.
My research interests sit at the meeting point of actuarial science, risk theory, dependence modelling, risk measures, optimization, statistical modelling, public health, climate, stochastic processes, and reinforcement learning. I use quantitative methods to understand patterns, measure uncertainty, and produce evidence that can support better choices.
Fig. 1A Markov chain, the model behind two of the studies below, running live.
Paused
day 1
| Dry | Light | Heavy | |
|---|---|---|---|
| Dry | 0.70 | 0.25 | 0.05 |
| Light | 0.40 | 0.40 | 0.20 |
| Heavy | 0.20 | 0.50 | 0.30 |
- Dry
- 100.0%
- Light rain
- 0.0%
- Heavy rain
- 0.0%
7 × 7
transition matrix
2019 – 2020
Rainfall in Kumasi: Markov Chains and Time Series
For my undergraduate research, I modelled and forecast rainfall in the Kumasi Metropolis using a first-order Markov chain and time-series methods. I constructed a 7 × 7 transition matrix to study how rainfall states changed over time and explored how statistical forecasting could support climate-related decisions.
In collaboration with Mr. Francis Doe and Dr. Kwame Piesie Frimpong.
First-order Markov chain · Time series · Forecasting
3 stations
of rainfall data, 2010–2023
2024 – 2025
Rainfall Data Availability: Hidden Markov Models
For my master’s research at the University of Ghana, I applied Hidden Markov Models to forecast rainfall data availability across Accra, Tamale, and Ho, using Ghana Meteorological Agency data from 2010–2023. I used the Viterbi and Baum–Welch algorithms to model hidden states and estimate parameters, then assessed model performance using AIC, BIC, and log-likelihood. The findings showed differences in station-level data reliability and demonstrated how HMMs can inform climate monitoring and resource management.
Conducted under the supervision of Professor F. O. Mettle.
Hidden Markov Models · Viterbi · Baum–Welch · AIC and BIC
16 regions
of Ghana, 2018–2022
2023 – 2024
Tuberculosis Across Ghana: Bayesian Spatial and Spatiotemporal Modelling
In a collaborative study, I contributed to a Bayesian framework examining tuberculosis prevalence and relative risk across Ghana’s 16 regions from 2018–2022. The analysis used Besag–York–Mollié (BYM) and Knorr–Held spatiotemporal models, implemented with INLA in R, to estimate regional risk, identify hotspots, assess spatial and temporal patterns, and examine public-health predictors including HIV prevalence, ART coverage, and treatment rates. The study found marked regional variation and persistent clustering, including elevated risk in Greater Accra, supporting more targeted public-health planning.
With Professor Dominic Otoo, Dr. Baaba A. Danquah, and Dr. Abdul-Karim Iddrisu.
BYM · Knorr–Held · INLA in R · Spatiotemporal
5,218
women’s records, 2022 Ghana DHS
2025 – 2026
Caesarean Delivery in Ghana: Statistics, Spatial Models, and Explainable Machine Learning
I recently worked on an analysis of caesarean-section delivery using 5,218 women’s records from the 2022 Ghana Demographic and Health Survey. The study combined survey-weighted logistic regression, Bayesian multilevel logistic regression, a BYM2 spatial model, and XGBoost interpreted with SHAP values. The results estimated caesarean-section prevalence at approximately 20.15%, identified education, wealth, multiple births, and maternal age among important predictors, and showed substantial community-level clustering. Spatial differences narrowed after adjustment for covariates, pointing to the importance of underlying social and structural conditions.
With Dr. Abdul-Karim Iddrisu.
Survey-weighted regression · Bayesian multilevel · BYM2 · XGBoost with SHAP