Reduced false positives by 40% for a Pan-African lender using an ensemble ML model on 2M+ transaction records.
Our client, a major FinTech lender processing 50,000+ daily loan applications, struggled with high false-positive fraud flags. We built a hybrid XGBoost + Neural Network ensemble trained on 5 years of transaction logs. We delivered a low-latency REST API (sub-100ms) integrated directly into their loan origination system, deployed on AWS with auto-scaling.
Project Manager
Team Members
- Amina Bello
- Chidi Okonkwo
- Kofi Mensah
Project Timeline
Data Audit & Feature Engineering
01 Jan 2026
Ingested 2M+ records and engineered 120 behavioural features.
Ensemble Model Training
15 Feb 2026
Trained XGBoost and DeepFF models; achieved 0.96 AUC on validation.
API & Integration
01 Mar 2026
Built FastAPI gateway with Redis caching for sub-100ms inference.
A/B Testing & Go-Live
20 Mar 2026
Ran A/B test on 20% traffic; rolled out to 100% after 1 week.
Challenges
Extreme class imbalance (0.5% fraud rate) while maintaining sub-100ms latency and full interpretability for compliance.
Results
40% reduction in false positives, 15% increase in borderline loan approvals, and 99.9% API uptime.
"electroSOFT turned our messy logs into a precision tool. We cut false positives by 40% in month one, saving our support team hundreds of hours." – CTO, Pan-African Lender
Project Files
Project Gallery
Interested in This Project?
Let's discuss how we can deliver similar results for your business. Book a free discovery call or reach out via WhatsApp.