Business Problem

👋 Welcome aboard

You're joining the Fraud Detection team at PayPal as a new data scientist. The team includes machine learning engineers, fraud analysts, and product managers who work together to protect millions of users from financial fraud.

✍️ Project Description

PayPal is one of the world’s leading digital payment platforms, enabling secure transactions for individuals and businesses globally. As a trusted intermediary for money movement, PayPal must ensure its users are protected from financial fraud at every step.

PayPal continues to face a critical challenge: fraudsters. These are individuals who sign up and use PayPal with the sole intent of stealing money from other users and funneling it into their own accounts.

Your mission is to design a machine learning system that flags potential fraudsters as early as possible. But this isn't just a technical problem — it’s a high-stakes balancing act:

• Mistakenly flagging a real customer can damage their trust and hurt PayPal’s reputation.
• Missing a fraudster can lead to real monetary losses and potential regulatory consequences.
• Every flagged case triggers manual review, so operational efficiency is critical.

🎯 Key Objectives

Build a data science solution that addresses the following:

1. Fraud Model: Build a model that predicts the likelihood of a transaction being fraudulent based on user and transaction features. You will be provided with a dataset containing historical user information (e.g., account creation date, country, KYC status) and transaction details (e.g., currency, amount, merchant category).
2. Operationalization: How will you utilize this model to catch fraudsters? If a fraudster is identified, what should be the resulting action: LOCK_USER, ALERT_AGENT, or BOTH?

• LOCK_USER - Current transaction is blocked and user’s account is LOCKED. This prevents the user from performing any transactions with their Revolut account. Access can only be restored after contacting an agent.
• ALERT_AGENT - Current transaction is not blocked. An alert is sent to a transaction monitoring agent for further review.

📊 Downloads

Boilerplate Template

To help you get started, we’ve provided a boilerplate in the form of a Jupyter Notebook.

Dataset

You’ll be working with a simulated dataset containing users and transaction data. Use these datasets to address the project’s key objectives using data science methods.