Academic work in operational research, machine learning for policy, and development economics.
This paper models the operational risk of misinformation on social media platforms using stochastic processes. A proof-of-concept Markovian framework captures state transitions between different levels across constructed belief risk and linguistic risk spectra, using a fine-tuned RoBERTa model, Claimbuster, and CardiffNLP's sentiment scoring model on real social media data. Monte Carlo simulation quantifies uncertainty in long-run misinformation prevalence under different platform intervention scenarios.
Accepted 3 March 2026 · Co-authored with and Supervised by Dr. Mehdi Toloo · Oral presentation · Preprint forthcomingUsing 2.6 million UK property records, this paper constructs a risk-scoring model for household fuel poverty using Markov Chain Monte Carlo Bayesian estimation and Random Forest classification. The paper effectively creates a triage between multiple Machine Learning approaches and a Bayesian inference. The model integrates energy performance certificate data, property characteristics, and local area socioeconomic indicators to produce small-area estimates of fuel poverty risk at a granularity unavailable from survey data alone, providing a sufficient level of validity through triage for operational purposes.
Co-authorship with Dr. Wolfgang Garn · Preprint forthcomingThis independent research applies Data Envelopment Analysis (DEA) to assess the efficiency of public infrastructure provision across Global South economies, examining the relationship between infrastructure investment, institutional quality, and output efficiency. The paper contributes to a literature on whether the constraints on infrastructure development in low-income contexts are primarily financial or institutional.
Preprint forthcoming