Hands-On Machine Learning for Payment Fraud Detection Training Course

Course Overview

Learning Objectives

• Frame payment fraud detection as a machine learning problem with clear labels and observation windows.
• Assess transaction, device and channel data for completeness, timeliness and point-in-time correctness.
• Engineer velocity, aggregation and behavioral features capturing deviations from normal customer activity.
• Build and compare supervised models, from logistic regression to gradient-boosted trees.
• Apply class weighting, resampling and cost-sensitive learning to severe class imbalance.
• Use unsupervised anomaly detection to surface new fraud patterns without historical labels.
• Apply graph analytics to expose shared devices, linked accounts and mule networks.
• Evaluate models with precision-recall, cost-based and operational measures instead of headline accuracy.
• Set score thresholds and decision strategies that reduce false positives within alert capacity.
• Design real-time scoring flows that meet latency limits and combine rules with models.
• Monitor deployed models for data drift, performance decay and fraudster adaptation.
• Produce explanations and reason codes that analysts, customers and reviewers can understand.
• Prepare model documentation and validation evidence for model risk management review.

Who Should Attend

Learning Outcomes

• By the end of this course, participants will be able to:
• Explain where machine learning adds value over rules in payment and transaction fraud detection.
• Define fraud labels, observation windows and time-based splits that prevent data leakage.
• Build velocity, aggregation and behavioral features from transaction, device and customer data.
• Train and compare logistic regression, tree-based and neural network fraud models.
• Handle severe class imbalance using weighting, resampling and cost-sensitive techniques.
• Detect emerging fraud patterns with unsupervised anomaly detection methods.
• Uncover fraud rings and mule networks using graph and network analytics.
• Evaluate model performance with precision-recall, value-based and operational measures.
• Set thresholds and decision strategies that reduce false positives and customer friction.
• Outline a real-time scoring design that meets latency and resilience requirements.
• Monitor production models for drift, performance decay and fraudster adaptation.
• Generate clear explanations and reason codes for fraud analysts and reviewers.
• Prepare a monitoring plan and validation pack aligned with model risk expectations.

Course Outline