FortiX: AI-Powered Fraud Detection Platform

1. Project Overview

FortiX is a web-based platform designed to analyze and detect fraudulent digital transactions. The system is built on a modern tech stack that separates the backend (the "brain") from the frontend (the "face").

The platform's workflow is built for batch analysis. A user uploads a dataset of new transactions, which are then scored against a pre-trained machine learning model. The results are then displayed in a rich, interactive dashboard with multiple pages for analysis.

2. Technology Stack

Backend (The "Brain")

• Framework: FastAPI. A high-performance Python framework for building the API.
• Server: Uvicorn. The server that runs the FastAPI application.
• Database: PostgreSQL. A powerful, professional-grade SQL database to store all transactions, alerts, and user feedback.
• ORM: SQLAlchemy (Async). The Python "translator" that lets our app talk to the PostgreSQL database using Python code instead of raw SQL.

Frontend (The "Face")

• Framework: React (with TypeScript/Vite). A modern JavaScript library for building interactive, component-based user interfaces.
• Routing: React Router (react-router-dom). Used to create a multi-page experience.
• Charts: Recharts. A declarative charting library for building visualizations.
• Package Manager: NPM (Node Package Manager). Used to install frontend libraries and run the development server.

Machine Learning (The "Intelligence")

• Library: Scikit-learn (sklearn). The most popular Python library for all-purpose machine learning.
• Data Handling: Pandas. Used to read, clean, and manipulate the transaction data.
• Model Storage: Joblib. Used to save our trained model to a file (fraud_model.joblib).

---

3. Project Pipeline (How it Works)

1. Startup: The FastAPI backend server starts and loads the pre-trained fraud_model.joblib into memory. It also connects to the PostgreSQL database.
2. Upload: A user visits the dashboard and uploads a test.csv file containing new transactions.
3. Analysis (Backend):
   • The backend receives this file at the POST /api/dataset/analyze endpoint.
   • It clears all old data from the Transaction and Alert tables to start a fresh analysis.
   • It reads the CSV and iterates through every single row.
   • For each transaction, it uses the loaded ML model to predict a risk_score (0-100).
   • It saves the transaction and its new risk score to the Transaction table.
   • If a transaction's score is high (e.g., > 70), it also creates a new record in the Alert table.
4. View (Frontend):
   • After the upload is complete, the user is redirected to the Analytics page.
   • This page (and all other pages) makes new API calls (e.g., GET /api/analytics/kpis) to fetch the analysis results that were just saved to the database.
   • The charts and tables are then populated with this data.

---

4. The Machine Learning Model

What Model is Used?

We are using a RandomForestClassifier.

Why This Model?

This was a deliberate choice for this project:

• It's Powerful: A Random Forest is an "ensemble" model, meaning it's a team of hundreds of small "decision trees" that all vote on the final answer. This makes it highly accurate.
• It's Good with Your Data: It's excellent at handling tabular (Excel-like) data and doesn't require as much fine-tuning as other models.
• It's Explainable: We can easily extract "feature importances" (e.g., "transaction amount was the most important factor"). This is critical for the "Risk Analysis" page.
• It's Balanced: It's the perfect balance of power and explainability, avoiding "black box" models like Neural Networks which are much harder to interpret.

Key Performance Metrics

(Note: These are example metrics. You should replace these with the actual metrics from your train_evaluate_model.py script's output.)

• Accuracy: 99.8%
• Precision (for "Fraud"): 0.88
• Recall (for "Fraud"): 0.91
• F1-Score (for "Fraud"): 0.89

---

5. How to Run the Project

1. Terminal 1 (Backend):

   • cd fortix-guard-engine
   • source .venv/Scripts/activate
   • cd fortix_backend
   • python -m uvicorn app.main:app --reload
   • (Leave this running)

2. Terminal 2 (Frontend):

   • cd path/to/frontend_folder
   • npm install (only run this once)
   • npm run dev (this starts the server)
   • (Leave this running)

3. Access: Open the http://localhost:5173 (or whatever URL npm gives you) in your browser.