Samsung EnnovateX 2025 AI Challenge Submission

Problem Statement - Classify User Application Traffic at the Network in a Multi-UE Connected Scenario

Team name - GadaSoftwares

Team members (Names) - Swayam Kohli (Leader), Mahak Singh, Archit Sahay, Maitreyi Jha

Demo Video Link - https://youtu.be/wdI0Zr8693M

---

Project Artefacts

Technical Documentation

All technical details are in the docs/ folder:

• approach.md: Solution approach and novelty
• architecture.md: System architecture and component flow
• installation.md: Step-by-step setup guide
• user_guide.md: Dashboard navigation and interaction
• features.md: Key features and screenshots

Source Code

The complete project source code is in the src/ folder:

• src/backend/: FastAPI server, ML inference, QoS engine
• src/frontend/: React + Vite dashboard, real-time UI, multi-UE monitoring

Models Used

None (we used only self-developed models)

Models Published

• CATO Multi-Class Traffic Classifier
  🤗 https://huggingface.co/ArcFR/Network_Traffic_Classifer/tree/main

Datasets Used

• Samsung_dataset.csv – Public dataset provided for training (used internally)
• website_testing.csv – Real-world encrypted traffic flows for testing
• 🔗 https://www.kaggle.com/datasets/jsrojas/ip-network-traffic-flows-labeled-with-87-apps
• 🔗 https://www.kaggle.com/datasets/kimdaegyeom/5g-traffic-datasets/data

Attribution

This project was built entirely from scratch during the hackathon. No existing open-source project was used as a base. All components — including the machine learning model, QoS engine, backend API, and frontend dashboard — were developed in-house by Team GadaSoftware.

The solution strictly uses:

• Open-source libraries (MIT/Apache 2.0 licensed): scikit-learn, FastAPI, React, Vite
• Public datasets: Samsung_dataset.csv, website_testing.csv
• Self-developed code and models

No third-party APIs, cloud services, or proprietary tools were used.