Fraud & Anomaly Detection on AWS

Business fraud is a growing concern across online and offline transactions. Fraud detection is one of the most important use-case for the Fintech startups. This session provides an overview of how you can use machine learning with Amazon SageMaker and Amazon Fraud Detector to implement a customized fraud detection and prevention solutions, how to proactively identify these use cases, and how to implement changes to protect your business and your customers.

Transcript

2. Rohini Gaonkar Senior Developer Advocate @rohini_gaonkar @rohinigaonkar
3. Fraud & Anomaly Detection on AWS with Amazon SageMaker and Amazon Fraud Detector
4. Fraud is everywhere
5. $22.8b 2016 $27.9b 2018 $32.9b 2021 $35.7b 2023 Data fromThe Nilson Report, November 2019, Issue 1164 (https://nilsonreport.com/upload/content_promo/The_Nilson_Report_Issue_1164.pdf ) Payment Fraud Trends
6. Customers want : Frictionless Payment experience Speedy transaction Processing Protection from fraudulent activity Merchants want : Quick decisions regarding consumers and their transactions Speedy transaction processing Accurate fraud detection and prevention Customers don’t want : False positives – incorrectly declined transactions Credit card fraud Merchants don’t want : False positives – the cost of resolving declined transactions Stolen payment credentials or account takeover fraud (ATO) Balancing the need for transaction speed with agile fraud protection Payments Fraud
7. Examples of Frauds in the Industry • Counterfeit Personal details • Identity Theft • Fake Bank Account • Synthetic fraudsters • Loan Phishing Scams • False Claims • Inflated Claims • False documents • Impersonation • Faked Death • Insurance Company Frauds Lending Insurance
8. Fraud detection is difficult $$$ billions lost to fraud each year Online business prone to fraud attacks Bad actors change tactics often Rules = more human reviews Dependent on others to update detection logic
9. The solution requires Data Analytics and Machine Learning at scale Authenticate customers Detect anomalies and threats Analyze data Apply machine learning Manage data Determine fraud score and continually improve As the landscape evolves and fraudsters improve their methods, the way to level the playing field is to analyze all available data—historical and real time—and apply Machine Learning to decipher legitimate transactions from illegitimate.
10. Solutions for Fraud Detection on AWS • Rule based • Machine Learning (Supervised & Unsupervised) • Tabular Data • Graph & Networks • AI Service - Amazon Fraud Detector
11. Typical Solution: Rule-based Fraud Detection DENY Over Limit High Rate Stolen Card ? APPROVE
12. Rule-Based Fraud Detection Challenges Static Rules Bug-Prone Complicated Cannot Scale Always Behind
13. Using Machine Learning Models for Fraud Detection Dynamic Self- improving Maintainable Scalable Real-time
14. Using Machine Learning Algorithms for Fraud Detection Two “flavors” of machine learning: • Supervised: Access to labeled data • Unsupervised: Access to features alone Problem: What to do when we don’t have annotated data, but want to identify potentially fraudulent transactions?
15. Using Machine Learning for Supervised Learning Feed labeled data to algorithm Discover relationships between input and output Apply solution to unseen data Make predictions
16. Using Machine Learning for Unsupervised Learning Feed raw data to algorithm Uncover hidden patterns Automatically flag anomalies Investigate potential fraud
17. AWS ML Stack VISION SPEECH TEXT SEARCH CHATBOTS PERSONALIZATION FORECASTING FRAUD DEVELOPMENT CONTACT CENTERS Ground Truth AWS Marketplace for ML Neo Augmented AI Built-in algorithms Notebooks Experiments Processing Model training & tuning Debugger Autopilot Model hosting Model Monitor Deep Learning AMIs & Containers GPUs & CPUs Elastic Inference Inferentia FPGA Amazon Rekognition Amazon Polly Amazon Transcribe +Medical Amazon Comprehend +Medical Amazon Translate Amazon Lex Amazon Personalize Amazon Forecast Amazon Fraud Detector Amazon CodeGuru AI SERVICES ML SERVICES ML FRAMEWORKS & INFRASTRUCTURE Amazon Textract Amazon Kendra Contact Lens For Amazon Connect SageMaker Studio IDE Amazon SageMaker DeepGraphLibrary
18. AWS ML Stack VISION SPEECH TEXT SEARCH CHATBOTS PERSONALIZATION FORECASTING FRAUD DEVELOPMENT CONTACT CENTERS Ground Truth AWS Marketplace for ML Neo Augmented AI Built-in algorithms Notebooks Experiments Processing Model training & tuning Debugger Autopilot Model hosting Model Monitor Deep Learning AMIs & Containers GPUs & CPUs Elastic Inference Inferentia FPGA Amazon Rekognition Amazon Polly Amazon Transcribe +Medical Amazon Comprehend +Medical Amazon Translate Amazon Lex Amazon Personalize Amazon Forecast Amazon Fraud Detector Amazon CodeGuru AI SERVICES ML SERVICES ML FRAMEWORKS & INFRASTRUCTURE Amazon Textract Amazon Kendra Contact Lens For Amazon Connect SageMaker Studio IDE Amazon SageMaker DeepGraphLibrary
19. Fully managed data processing jobs and data labeling workflows One-click collaborative notebooks and built- in, high performance algorithms and models One-click training Debugging and optimization One-click deployment and autoscaling Amazon SageMaker helps you build, train, and deploy models Visually track and compare experiments Automatically spot concept drift Fully managed with auto-scaling for 75% less Prepare Build Train & Tune Deploy & Manage 101011010 010101010 000011110 Collect and prepare training data Choose or bring your own ML algorithm Set up and manage environments for training Train, debug, and tune models Deploy model in production Manage training runs Monitor models Validate predictions Scale and manage the production environment Add human review of predictions Web-based IDE for machine learning Automatically build and train models CI/CD
20. Why use SageMaker for Fraud Detection? • Use a notebook for model development • Easily scale to full dataset later. • One click Training and Deployment • Built-in Algorithms • Use your own custom model, with full control.
21. Our Solution Train XGBoost Model using SageMaker* Train Random Cut Forest Model using SageMaker* Labeled Data Unlabeled Data Deploy XGBoost Model Deploy Random Cut Forest Model Live Data e.g. incoming, real-time transactions Predictions e.g. Anomalous transactions, fraud * SageMaker built-in algorithm
22. Optional Anomaly Detection Fraud Detection Solution Architecture Amazon API Gateway AWS Lambda Amazon SageMaker (XGBoost) Amazon SageMaker (Random Cut Forest) Amazon S3 bucket (Model and Data) Amazon S3 bucket (Results) Amazon QuickSight Amazon Kinesis Data Firehose Transactions
23. Demo
24. Getting Started Visit the getting started guide Documentation > Access code & use sample data set or customize to deploy the solution Code on GitHub >
25. Solutions for Fraud Detection on AWS • Rule based • Machine Learning (Supervised & Unsupervised) • Tabular Data • Graph & Networks • AI Service - Amazon Fraud Detector
26. Fraud Detection on Graphs and Networks
27. Graph Data Use Cases Node Edge Homogeneous Heterogeneous Heterogenous Financial Transaction Network
28. Fraud Detection with Graphs Fraud can evolve and adapt to fool rules-based or simple features-based methods Fraudsters cannot mask their behavior with respect to the full interaction graph Node aggregation : suspicious accounts connect with too many other entities Activity aggregation : suspicious accounts begin to act in tandem
29. Deep Graph Library (DGL) GNN Models • GCN • GAT • R-GCN • GraphSage • … DGL-LifeSci DGL-KG Graph NN Modules GNN Message-Passing Interface Graph Algorithm DGL Runtime PyTorch MXNet TensorFlow GPU(s) CPU Cluster Deep Graph Library Backend Platform … …
30. Fraud Detection with Graph Neural Networks
31. Deep Graph Library on Amazon SageMaker Scalable Fast performance Fully managed
32. Solution Architecture S3 Bucket Data, Models, Results Data Preprocessing Lambda function Amazon SageMaker Training Amazon SageMaker Processing Amazon SageMaker Notebook Manual Orchestration Model Training Lambda function Amazon SNS Notify of predicted fraud accounts Amazon DynamoDB Store user/account classification Postprocessing Lambda Function
33. Access code & use sample data set or customize to deploy the solution Code on GitHub > Detecting fraud in heterogeneous networks using Amazon SageMaker and Deep Graph Library Blog on AWS Machine Learning Blog > Fraud Detection in Financial Transaction Networks with Amazon SageMaker – Webinar YouTube video demo Getting Started
34. Key Takeaways • Machine learning allows us to adapt to fraudsters’ behavior • Unsupervised learning provides insights even when we don’t have labels • For data in the form of transaction networks and graphs, the SageMaker Graph Fraud Detection solution provides a specialized graph neural network model. • The SageMaker solutions for fraud detection are great starting points, with all necessary parts included to quickly deploy models in production.
35. Fraud detection is difficult $$$ billions lost to fraud each year Online business prone to fraud attacks Bad actors change tactics often Rules = more human reviews Dependent on others to update detection logic
36. Fraud detection with ML is also difficult Top data scientists are costly & hard to find One-size-fits-all models underperform Often need to supplement data Data transformation + feature engineering Fraud imbalance = needle in a haystack
37. The AWS ML Stack Broadest and most complete set of Machine Learning capabilities VISION SPEECH TEXT SEARCH CHATBOTS PERSONALIZATION FORECASTING FRAUD DEVELOPMENT CONTACT CENTERS Ground Truth AWS Marketplace for ML Neo Augmented AI Built-in algorithms Notebooks Experiments Processing Model training & tuning Debugger Autopilot Model hosting Model Monitor Deep Learning AMIs & Containers GPUs & CPUs Elastic Inference Inferentia FPGA Amazon Rekognition Amazon Polly Amazon Transcribe +Medical Amazon Comprehend +Medical Amazon Translate Amazon Lex Amazon Personalize Amazon Forecast Amazon Fraud Detector Amazon CodeGuru AI SERVICES ML SERVICES ML FRAMEWORKS & INFRASTRUCTURE Amazon Textract Amazon Kendra Contact Lens For Amazon Connect SageMaker Studio IDE Amazon SageMaker DeepGraphLibrary
38. Amazon Fraud Detector A fraud detection service that makes it easy for businesses to use machine learning to detect online fraud in real-time, at scale.
39. Detect common types of online fraud • New account fraud, within an account sign-up process • Online payment • Guest checkout fraud • Promotion and loyalty program abuse • Online identity fraud • Loyalty account protection Easily identify potentially fraudulent online activities
40. Key features Pre-built fraud detection model templates Automatic creation of custom fraud detection models Models learn from past attempts to defraud Amazon Amazon SageMaker integration Interface to review past events and detection logic
41. How it works https://aws.amazon.com/fraud-detector/
42. ML Template: Automated model building 1 2 4 5 Training data in Amazon S3 6 3
43. How it works https://aws.amazon.com/fraud-detector/ Amazon Fraud Detector -
44. Generating Fraud Predictions Guest Checkout: Purchase IP: 1.23.123.123 email: joe@example.com Payment: Bank123 … Fraud Detector returns: Outcome: Approved ML Score: 160 Purchase Approved Call service with: IP: 1.23.123.123 email: joe@example.com Payment: Bank123 … AWS Cloud
45. Demo
46. 3rd Party blog -> Towards Data Science - Understanding AUC - ROC Curve AUC - Area Under The Curve TP – True Positives TN – True Negatives FP – False Positives FN – False Negatives
47. Getting Started… Visit the getting started guide Documentation> Access samples Code on GitHub >
48. Solutions for Fraud Detection on AWS • Rule based • Machine Learning [ GitHub awslabs ] • Tabular Data • Graph & Networks • AI Service - Amazon Fraud Detector [ GitHub aws-samples ]
49. Thank You! @rohini_gaonkar @rohinigaonkar Slides available on SlideShare & SpeakerDeck @rohinigaonkar