Intelligent System for Bank Payment Fraud Detection Based on Machine Learning

This project demonstrates how to build an end-to-end machine learning pipeline using Scikit-learn to achieve real-time prediction and visual display of fraud risks in bank transactions, including a complete front-end and back-end architecture and a PostgreSQL data layer.

With the increasing popularity of digital payments today, financial fraud has become a major threat to global financial institutions. According to statistics, the annual losses caused by payment fraud amount to billions of US dollars, and fraud methods are constantly evolving—from traditional credit card theft to complex identity theft and social engineering attacks. Traditional rule-based detection systems are no longer able to cope with these sophisticated threats. The rise of machine learning technology has brought new hope to the field of financial security: by analyzing massive historical transaction data and identifying hidden patterns and abnormal behaviors, machine learning models can issue early warnings before fraud occurs. The fraud-detection-intelligence-system project introduced in this article is exactly a complete practical case showing how to apply machine learning to bank payment fraud detection.

This project is an end-to-end intelligent fraud detection system designed to demonstrate how to use machine learning technology to enhance the security of financial transactions. The system has the following core functional features:

Interactive Transaction Data Input: Users can input transaction details through an intuitive web interface, including multi-dimensional data such as transaction amount, time, merchant information, and geographic location.

Machine Learning Risk Prediction: The system has a built-in trained machine learning model that can calculate the fraud probability of each transaction in real time, providing a quantitative basis for decision-making.

Instant Result Feedback: Risk assessment results are immediately displayed on the user interface, including fraud probability percentage and risk level prompts, helping users quickly judge transaction security.

Backend API Service: Adopts a modern API architecture, responsible for coordinating data flow between the front-end interface and the machine learning model, ensuring high availability and scalability of the system.

The project adopts a clear layered architecture design with well-defined responsibilities for each layer, facilitating maintenance and expansion:

The system uses PostgreSQL as a relational database to store three types of core data:

• User Data: Contains user registration information, login credentials, and permission configurations
• Transaction Data: Records detailed information of each transaction, serving as model input and historical archives
• Prediction Results: Stores the output of each risk assessment, supporting subsequent auditing and analysis

For the local development environment, the project provides a Docker Compose configuration; developers only need to run docker compose up -d to quickly start the database service.

The core of fraud detection is the machine learning model; the project uses the mature Scikit-learn library to build a pipeline-based machine learning system:

• Pandas: Used for data cleaning, feature engineering, and format conversion
• NumPy: Provides efficient numerical computation support
• Scikit-learn: Builds a complete machine learning pipeline, including data preprocessing, feature scaling, model training, and prediction
• Joblib: Used for serializing and storing models, ensuring that models can be deployed persistently

The advantage of this pipeline design pattern is that it can encapsulate data preprocessing and model inference steps into a single unit, avoiding data processing inconsistencies between the training and inference phases.

The backend is developed using the Python FastAPI framework, a modern, high-performance asynchronous web framework with the following advantages:

• Automatic API Documentation: Automatically generates interactive API documentation based on the OpenAPI specification
• Type Hint Support: Makes full use of Python's type annotations to improve code maintainability
• Asynchronous Processing Capability: Based on Starlette and Pydantic, supports high-concurrency request processing
• Data Validation: Automated request data validation and serialization

The backend service is responsible for receiving transaction data from the frontend, calling the machine learning model for prediction, and returning the results to the frontend for display.

The frontend uses the classic three-piece technology stack:

• HTML5: Semantic page structure
• CSS3: Modern style design and responsive layout
• JavaScript: Dynamic interaction logic and data visualization

Users access the system through a browser and can perform operations such as entering transaction data, viewing fraud detection results, and browsing historical prediction records.