MediGuard-AI: An Intelligent Audit System for Medical Insurance Claims Based on OCR and Machine Learning

MediGuard-AI is an end-to-end AI-driven medical insurance claim processing system that combines Tesseract OCR, Random Forest classifier, and modern web technologies to enable automatic medical document recognition, fraud detection, and claim decision-making. This system aims to address the pain points of traditional claim processes such as being cumbersome, error-prone, high fraud losses, and long claim cycles, and improve audit efficiency and accuracy through automated processes.

Traditional medical insurance claim processes rely heavily on manual audits, from receiving documents to verifying terms and making decisions, which is time-consuming and labor-intensive, and can easily lead to fraudulent claims being approved or legitimate claims being rejected due to human negligence. The global medical insurance industry loses tens of billions of dollars annually due to fraud, and patients often complain about long claim cycles. How to improve efficiency while ensuring accuracy is a common challenge for the industry.

MediGuard-AI is an open-source semester project developed by Srijan0409, using a full-stack architecture that integrates front-end interaction, back-end services, OCR, and machine learning. Its core goals are: users upload medical documents in PDF/image format, the system automatically extracts key information, calculates fraud probability scores, and provides "approve" or "reject" decision recommendations, with the entire process requiring no manual intervention and a response time in seconds.

The system consists of five modules:

1. Front-end: Built with pure HTML/CSS/JS, lightweight and easy to deploy, with file upload via Fetch API.
2. Back-end: Node.js + Express framework, responsible for routing and business logic, calling OCR and Python ML subprocesses.
3. OCR engine: Tesseract.js, supporting over 100 languages, recognizing printed text, handwriting, and table text.
4. ML module: Python + Scikit-Learn's Random Forest classifier, trained to generate .pkl models and return fraud probability predictions.
5. Data persistence: MongoDB + Mongoose, saving upload metadata, text, prediction results, etc., for auditing and iteration.

The system adopts a hybrid strategy of rules + machine learning: OCR extracts high-risk keywords (such as "emergency", "surgery", "altered") from the text, and their frequency and context are used as features input into the Random Forest model. The rule engine quickly marks obviously suspicious cases, while ML captures subtle fraud patterns; the two complement each other to improve detection effectiveness.

Deployment requires installing Node.js (v14+), Python 3.9+, and MongoDB Community Edition. The steps are as follows:

1. Clone the project code
2. Install Python dependencies: pip install -r ml-model/requirements.txt
3. Train the model: python ml-model/train.py
4. Install Node dependencies: npm install
5. Start the service: npm start
6. Visit http://localhost:5000 When using, upload test documents, and the system will display the fraud probability percentage and final decision in real time.

As a semester project, it has the following limitations: OCR accuracy is limited for complex layouts/low-quality scanned documents; Random Forest performs less well than deep learning on large-scale high-dimensional data; security design (user authentication, input validation) is relatively simple. Improvement directions: Introduce LayoutLM to enhance OCR; use Wide&Deep instead of Random Forest; add multi-language support and improve audit logs.

MediGuard-AI is an excellent entry-level AI application project that fully demonstrates the process of integrating OCR, traditional machine learning, and web technologies to solve practical business problems. For developers who want to understand the application of AI in fintech, it is a good learning case covering the entire process from data preprocessing to model deployment, with high educational value.