Praise Amonye
Speaker at AMLD Africa 2026
Praise Amonye is an emerging thought leader at the intersection of mathematics, artificial intelligence, and financial engineering, applying advanced quantitative methods to address complex global challenges. He is the Founder of Deep Afrika, a research and innovation initiative focused on developing data-driven solutions for Africa’s most pressing development and health needs. With a strong academic foundation in pure and applied mathematics and advanced training in financial engineering, he has led high-impact projects in financial risk modeling, machine learning, and healthcare innovation. His work explores how optimization techniques and predictive analytics can strengthen financial systems and improve public health outcomes. Praise has co-organized and spoken at international conferences and is widely recognized for his ability to bridge deep technical expertise with strategic insight. Driven by a passion for innovation for development, he is committed to advancing the future of technology, project management, and health through research-led practice.
At AMLD Africa
WITS University, Johannesburg, South Africa
COST-AWARE CREDIT SCORING FOR AFRICAN MICROFINANCE
In this presentation, I share my work on cost-aware and interpretable machine learning for credit scoring in African microfinance, grounded in my research at the intersection of mathematics, AI, and African financial systems. Through hands-on experience with data-driven initiatives across West Africa, I approach lending decisions not as abstract classification problems, but as choices with real consequences for institutions, borrowers, and local economies. Using a representative West African microfinance dataset of 10,000 loan records, I address a key challenge faced by many MFIs: making reliable lending decisions under limited data, strong class imbalance, and asymmetric costs of error. Rather than pursuing model complexity for its own sake, this work prioritizes economic relevance, presenting a production-aware machine learning pipeline designed to minimize expected financial loss rather than optimize accuracy alone. The pipeline includes missing value imputation, feature engineering and scaling, imbalance handling with SMOTE, feature selection via ANOVA F-tests, model comparison, and cost-sensitive threshold optimization. I evaluate several models and show that a simple, interpretable logistic regression model outperforms more complex alternatives, achieving an AUC of 0.603 on a held-out test set. By optimizing decisions using an expected loss framework, the model delivers an estimated 35% reduction in expected portfolio loss. SHAP-based interpretability supports transparency, regulatory compliance, and trust among loan officers. I conclude with a clear pathway to real-world adoption through pilot partnerships, live A/B testing, and model scaling, aligned with my mission to advance context-aware, transparent AI for African financial systems.
