Programme Parameters
Graduate / Intermediate
3 Days (4.5 Hrs)
Online (e-LMS)
• A basic understanding of the subject area and fundamental programming or scientific concepts.
• A laptop or desktop with a stable internet connection.
• Willingness to complete assignments and the capstone project.
About This Course
Fraud Detection Using AI in Finance teaches how machine learning finds the rare, adversarial needle in the transactional haystack. You build models for the core problem — flagging fraudulent transactions, accounts and behaviours — using both supervised methods on labelled fraud and unsupervised anomaly detection for the unknown. The course tackles what makes fraud hard: extreme class imbalance, adversaries who adapt, the cost trade-off between missed fraud and false alarms, and the need to score in real time. You finish able to build and evaluate a defensible fraud-detection model. A verified e-Certificate of competency and e-Marksheet from the Deep Science & Technology Consortium.
🎯 Program Aim
This course applies AI to financial fraud detection — anomaly detection, supervised and unsupervised models, imbalanced data and real-time scoring for payments and transactions.
📋 Course Objectives
1. Build supervised fraud-classification models.
2. Apply anomaly detection for unknown fraud.
3. Handle extreme class imbalance.
4. Balance missed fraud against false positives.
5. Design for real-time transaction scoring.
👥 Who Should Enroll?
• Fraud, risk and financial-crime analysts
• Data scientists in banking and fintech
• Payments and security professionals
• Students of financial analytics
🚀 Key Learning Outcomes
• The ability to build a fraud-detection model.
• A financial-fraud analytics project.
• A cost- and adversary-aware approach.
• A verified e-Certificate of competency and e-Marksheet from the Deep Science & Technology Consortium.
💎 What You'll Gain
Live & Recorded Sessions
Lifetime access to class recordingse-Certificate on Completion
Cryptographically verified credentialPost-Programme Support
Direct access to mentors & councilHands-On Experience
Notebooks, real-world code & datasetsCurriculum Outline
Fraud as a Modelling Problem
• Extreme class imbalance and why accuracy is a meaningless metric here
• Adversarial drift: fraudsters respond to your model in a way churners do not
• Label delay and the chargeback window that means today's labels are incomplete
Behavioural and Network Features
• Velocity, recency and aggregation features over sliding windows
• Entity resolution and device or identity graphs
• Graph features: shared attributes, community detection and ring identification
• Preventing leakage from features computed after the event
Supervised, Unsupervised and Hybrid Detection
• Gradient-boosted trees as the practical baseline and why they are hard to beat
• Anomaly detection for novel fraud with no labels yet
• Graph neural networks for organised fraud rings
• Ensembling with rules engines rather than replacing them
Thresholds, Cost and Customer Friction
• Cost-sensitive thresholds: false-positive friction against fraud loss
• Precision at low alert volumes, because analyst capacity is finite
• Champion/challenger deployment and measuring real uplift
Explainability and Regulatory Expectations
• SHAP-based reason codes for adverse decisions
• Fairness testing where a false positive freezes a customer's money
• Model documentation and audit trails for financial supervisors
Technical Specifications
| Parameter | Requirement |
|---|---|
| Covered Tool / Platform | Python |
| Covered Tool / Platform | TensorFlow |
| Covered Tool / Platform | PyTorch |
| Covered Tool / Platform | Keras |
| Covered Tool / Platform | Scikit-learn |
| Covered Tool / Platform | Jupyter Notebook |
| Covered Tool / Platform | Google Colab |
| Covered Tool / Platform | Hugging Face |
Frequently Asked Questions
Scholar Feedback & Reviews
Based on 0 scholar submissions
Rating Breakdown
No verified reviews published yet. Be the first to share your academic experience.
Leave Scholar Feedback
Your rating will help prospective scholars. Ratings below 3 stars are routed privately to the faculty mentor for immediate response.
