An Integrated Cybersecurity Strategy

Based on Okoye O. Kelechi's MSc project (University of Jos, 2025) on banks and fintechs in Nigeria

The problem

Debit card fraud through skimming, card-not-present attacks, SIM swaps and social engineering is rising across Nigerian ATM, POS and online channels. Using MFA alone adds friction to every payment, and RBA alone can be fooled by a single weak signal. Rule-based systems also miss new fraud patterns, and labelled fraud data is rare.

The solution

Unsupervised anomaly detection learns what normal looks like for each cardholder. An ensemble of an Autoencoder and an Isolation Forest produces one risk score, and that score decides how much authentication to ask for. Real customers get a smooth experience and fraudsters meet resistance.

Model flow

  1. 1Transaction
    ATM, POS, web, USSD or mobile request from a debit card
  2. 2Feature extraction
    Amount deviation, device, location, time, velocity, channel and merchant
  3. 3Autoencoder + Isolation Forest
    Two unsupervised models score the anomaly with no labelled fraud data needed
  4. 4Ensemble risk score
    A weighted fusion lowers false positives compared to either model alone
  5. 5Risk-Based Authentication
    Allow · OTP · Biometric + PIN · Block
  6. 6Analyst feedback
    Confirmed cases help tune weights and thresholds over time

Note: this demo runs simplified, deterministic versions of both models in the browser, using demo cardholders.