Machine Learning-Based Detection of Illegal Currency Trading in Zimbabwe’s Banking Sector

Illegal currency trading, facilitated through formal banking channels, has exacerbated Zimbabwe's economic instability, contributing to hyperinflation (175.8% in 2023) and eroding trust in financial institutions (ZimStat, 2023; Tsarwe & Mare, 2021) ([30],[10]). This study develops and evaluates a machine learning (ML) framework to detect illicit transactions in real-time, addressing critical gaps in Zimbabwe's reactive surveillance infrastructure. Using a dataset of 50,000 anonymized transactions (2020–2023) from three Zimbabwean banks—15% labelled as suspicious via RBZ audits—the research implemented Random Forest (RF), Support Vector Machines (SVM), and XGBoost algorithms. Feature engineering, guided by the Fraud Triangle Theory, identified key indicators: exchange rate variance (deviations >120% from official rates), transaction velocity (>15/hour), and network clusters with blacklisted entities. XGBoost emerged as the optimal model, achieving 95% precision, 91% recall, and 0.97 AUC-ROC, outperforming RF (93% precision) and SVM (84% F1-score).