README.md

metadata

title: my-backend
sdk: docker
app_port: 7860

FuseChain: Multimodal Ethereum Scam Classification System

FuseChain is a supervised scam classification system that fuses on-chain Ethereum transaction features with off-chain contextual signals market dynamics, Reddit sentiment, and Twitter engagement to classify EOA addresses as scam or normal.

Unlike conventional on-chain-only approaches, FuseChain operates at the address level, aggregating a behavioural profile across each EOA's full transaction history and the prevailing off-chain conditions during its operational period.

Project Architecture

This repository operates on three decoupled components:

1. Backend (FastAPI): Serves the XGBoost classifier via REST API endpoints, with pre-loaded feature caches for zero-latency inference and SHAP-based explainability.
2. Frontend (React + Tailwind): Interactive dashboard for analysts to visualise risk scores, SHAP feature contributions, narrative explanations, and batch analysis results.
3. ML Pipeline: Seven-notebook pipeline that processes raw data through temporal alignment, address-level aggregation, SHAP feature selection, and model training.

Setup Instructions

1. Backend Setup

cd backend
python -m venv venv
# Activate venv:
# Windows: venv\Scripts\activate
# Mac/Linux: source venv/bin/activate

pip install -r requirements.txt
python run.py
# Server runs at http://localhost:8000
# API Docs at http://localhost:8000/docs

2. Frontend Setup

cd frontend
npm install
npm run dev
# Dashboard runs at http://localhost:5173

3. ML Environment

cd ml_pipeline
pip install -r requirements.txt
jupyter lab

Dataset

The FuseChain dataset used to train this model is publicly available on Hugging Face:

FuseChain Multimodal Ethereum Fraud Dataset

Model

The FuseChain model is publicly available on Hugging Face:

FuseChain Model Repository