DeFi Users Behavior Clustering

This project aims to analyze and cluster user behavior in decentralized finance (DeFi) platforms using open-source data. By examining user interactions, transaction patterns, and other relevant metrics, the goal is to uncover meaningful insights into how users engage with DeFi applications. These insights can help to improve platform designs, identify emerging trends, and provide valuable information for both developers and users within the DeFi ecosystem.

---

Scope of Analysis_

The scope of this analysis encompasses various decentralized finance (DeFi) protocols and platforms.
Specifically, the study focuses on the following types of DeFi protocols on the Ethereum blockchain:

• Decentralized Exchanges (DEX): Uniswap, Curve DAO, Balancer
• Lending Platforms: Aave, Maker
• Stablecoins: Tether, USD Coin (USDC), Dai
• Yield Farming: Yearn Finance, Harvest Finance
• Non-Fungible Token (NFT): NFTfi

---

Data Collection_

The data used in this analysis is sourced from the Ethereum blockchain and all the data collection process is detailed in the section Data Collection_ of the HF Space.
Process include information on protocols, user transactions and average usages, protocol interactions or market metrics.

Timeframe_ (2 years: 2023-2024)

• Start - 31st of December 2022, 22:59:59 UTC
• End - 30th of December 2024, 23:00:11 UTC

Metrics_

• Total protocols (see scope above): 11
• Total unique transactions : 22 682 739
• Total unique users/addresses : 6 876 845
• Total market hours covered : 177 955

Storage_

All data are stored in Parquet format and are available in the HF Dataset.

├── contracts.parquet         # Contains contract details for selected DeFi protocols.
├── transactions.parquet      # Contains transaction data for Ethereum-based contracts.
├── market.parquet            # Contains enriched market data with aggregated transaction metrics.
└── users.parquet             # User profiles based on transaction data.

---

Trends Analysis_

By analyzing transactions, we can identify trends and correlation patterns within the DeFi ecosystem. All the trends analysis is available in the section Trends Analysis_ of the HF Space.

Correlation_

With graphs below, we can conclude that there is a strong correlation between the number of users and the numbers of transactions.
This also signify that the number of transactions per user is approximately repetitive & stable.

Transactions Number vs Users per Protocol

Transactions Number vs Users per Type

Proportion Trends_

Here, we can observe that stablecoins represent the major usage.
Also, in termes of trends, the global count of user and the proportion of DEX and Lending protocols usage is increasing over time.

Unique Transactions per Protocol by Month

Unique Transactions per Type by Month

---

Features Engineering_

The features generated for each user address are detailed in the section Feature Engineering_ of the HF Space.
Process include the following steps and allows to obtain a total of 62 features for each user address:

1. Loading & Processing Data (from .parquet files)
2. Aggregating User Metrics, Transactions Data, Market Data
3. Standardizing Features (with a specific method described in step 6 of the section)

Storage_

The features files are saved in arrow format and are also available in the HF Dataset.

├── features.arrow                   # Contains the 62 features generated for each user address.
└── features_standardised.arrow      # Contains the 62 features standardized following the process detailed in step 6.

---

Clustering_

By identifying distinct clusters, we can gain insights into different user profiles and behaviors within the DeFi ecosystem. To perform the clustering analysis, K-means algorithm is used to group users into clusters based on their features and transactional activities.

Process_

1. PCA: reduce dimensions. Here, 6 sigmas of the variance is explained by 28 dimensions against 62 initially.
2. Elbow Method: analyze visually the optimal number of clusters for the K-means algorithm.
3. Hyperparameters Tuning: identify the optimal hyperparameters with Optuna.

Results_

The analysis of main differences between each clusters is available in the section Clustering_of the HF Space. The clusters are described with their characteristics, interactions types, correlation matrix, transactions activity, diversity and influence, sent and received transactions statistics, exposure metrics and timing behavior.

---

Performance Report_

As the final step of the project, the performance report delivers insights into a user's behavioral analysis. It includes a summary of the cluster analysis associated with the user, an evaluation of strengths and weaknesses within the DeFi ecosystem, and offers recommendations to enhance performance.

Text-to-Text Reporting_

A text-to-text report is generated with llama-3.3-70b-versatile.
The report includes an analysis of strengths, areas for improvement, and actionable recommendations to enhance performance.

Metrics_

Metrics are displayed for global and cluster ranks. The radar charts provide a visual representation of the user's performance across various metrics, highlighting strengths and areas for improvement.

Note

Experience available in the section Performance Report_ of the HF Space.

---

Run the App Locally

Prerequisites

Before you begin, ensure you have the following software installed on your system:

• Python 3.12+: This project uses Python, so you'll need to have Python installed. You can download it from python.org.
• uv: This project uses uv for dependency management. Install it by following the instructions at docs.astral.sh/uv.
• Docker (Optional): If you prefer to run the project in a Docker container, ensure Docker is installed. Instructions can be found at docker.com.

---

Installation

Follow these steps to install and set up the project:

1. Clone the Repository:

   git clone https://github.com/mriusero/defi-user-behavior-clustering    # Clone the repository
   cd defi-user-behavior-clustering                                       # Access the project directory

2. Create Virtual Environment:

   • On MacOS/Linux:

     python -m venv .venv         # Create a virtual environment
     source .venv/bin/activate    # Activate the virtual environment

   • On Windows:

     python -m venv .venv         # Create a virtual environment
     .venv\Scripts\activate       # Activate the virtual environment

3. Install Dependencies:

   pip install uv               # Install uv for dependency management
   uv sync                      # Install dependencies with uv

   This will install dependencies listed in pyproject.toml.

---

Running Locally

To run the application locally, execute the following command:

 uv run streamlit run app.py

This will start the application, and you can access it in your web browser at http://localhost:8501.

Running Locally with Docker

1. Build the Docker Image:

   docker build -t defi-behavior-app .

2. Run the Docker Container:

   docker run -p 8501:8501 defi-behavior-app

This will start the application, and you can access it in your web browser at http://localhost:8501.

---

License

This project is licensed under the terms of the MIT License. If you find this project helpful, please consider giving it a star or citing it in your work.