SQL-Based Fraud Detection Using PaySim Dataset

This project uses SQL to analyze a synthetic mobile money transaction dataset and explore fraud detection challenges.

Project Overview

Data Source

• PaySim Dataset on Kaggle
• Note: For this project, only the first 10,000 rows of the dataset were used to simplify analysis and improve performance within Microsoft SQL Server.

Folder Structure

File	Description
schema.sql	Creates users, transactions, transaction_types tables
seed.sql	Populates tables from raw PaySim data
queries.sql	Exploratory queries and fraud detection logic

Schema Design

• users: all unique user IDs from senders/receivers
• transaction_types: unique transaction types (e.g. TRANSFER, CASH_OUT)
• transactions: main table linking users, types, and fraud indicators

Key Insights

• Out of 10,000 transactions, 68 were marked as real fraud (isFraud = 1)
• The built-in fraud detection rule (isFlaggedFraud) failed to detect any of them (0% success)
• Fraudsters often use TRANSFER and CASH_OUT types

Limitations

• Dataset is synthetic and simplified (10k rows used from full 6M+)
• Balance columns can't be used to detect fraud

Future Work

• Apply this logic to full dataset for deeper patterns
• Build custom fraud flagging logic in SQL
• Create dashboards using Power BI

Source

• PaySim Dataset on Kaggle

---