Project Summary:

This project focuses on building an anomaly detection system to help a bank identify potentially fraudulent transactions and monitor customers' monthly spending. By leveraging an unsupervised learning approach, particularly Isolation Forest, we aim to detect unusual transactions without needing historical fraud labels, a key advantage in identifying emerging fraud patterns that a supervised model might miss.

Project Objectives:

• Anomalous Transaction Detection: Develop a model to flag transactions that deviate from typical spending patterns. We experimented with several algorithms, including K-Means, DBSCAN, and One-Class SVM. However, Isolation Forest was ultimately chosen for its efficiency with large datasets, robustness in detecting outliers, minimal parameter tuning, and ability to provide an interpretable anomaly score. This approach enables proactive fraud detection by identifying outliers in real time.

• Monthly Limit Monitoring: Create a function that checks if any user has exceeded their monthly spending limit. This function filters transactions by user and date, then compares the total spending to the credit limit. It can run daily, giving the bank a continuous view of customers’ spending to prevent overspending.

Data Used:

• cc_info.csv: Contains credit card details for 843 users, including anonymized credit card numbers, spending limits, and user location information.

• tx_data.csv: Holds 230,000 transaction records, with details on transaction amounts, dates, and geographical locations.

Outcome: The system allows the bank to identify and investigate potential fraud and over-limit spending proactively, protecting customers and enhancing financial oversight. By integrating anomaly detection with daily spending checks, the project aims to deliver a comprehensive fraud monitoring tool adaptable to new fraud trends and spending behaviors.