Implement real-time fraud detection pipeline with anomaly and pattern…

[Cedarich]

📌 Description

This PR introduces a production-ready fraud detection module with the following capabilities:

• 🛠 Feature Pipeline

  • Normalized, velocity, geo, and time-based features

• 🤖 Anomaly Ensemble

  • Isolation Forest + One-Class SVM approximations

• 📊 Pattern Rules

  • Rapid transaction bursts
  • High-value in new geos
  • Card-not-present spikes
  • Merchant activity anomalies
  • Account takeover signals
  • Cash-out sequences

• ⚖ Risk Scoring

  • Interpretable reasons & configurable thresholds

• ⚡ Real-Time SLA Guard

  • <100ms response time
  • Latency tracking + sensitivity adjustment

• 🔁 Feedback Loop

  • Reduce false positives
  • Improve recall over time

• 📈 Monitoring Metrics

  • Totals, frauds, false positives
  • Average latency

---

🧩 Integration

• Exposes AIService.detectFraud for simple consumption.
• Wired exports available via ai/index.ts.

---

✅ Acceptance Criteria

• Latency under SLA (<100ms)
• High precision + broad fraud pattern coverage
• Interpretable + tunable scoring system
• Real-time monitoring + scalable architecture

---

🧪 Testing

1. Build the core module
2. Run the demo script
3. Validate:
   • Latency
   • Risk categorization results
   • Calibration via feedback loop

---

🚀 Impact

Enables immediate fraud risk scoring across apps using @chenaikit/core with clear, tunable, and monitored behavior.

close #31