Crypto RugPool Finder

Inspiration

The inspiration for this project stemmed from the increasing number of cryptocurrency scams like $SQUID, and $HAWK, specifically "rug pulls," where developers abandon a project after acquiring substantial investment.

What it does

The Crypto Rug Pull Finder is designed to help users analyze the legitimacy of cryptocurrency tokens by synthesizing data from multiple sources, performing sentiment analysis, and leveraging machine learning to predict potential rug pulls. The platform aims to provide real-time insights into token liquidity, volume, market capitalization, and social sentiment.

How we built it

Data collection involved extracting posts and comments from the Reddit API, retrieving market data from CoinGecko API, and accessing on-chain transaction data from the Moralis API. For sentiment analysis, we applied VADER to determine whether discussions about a token were positive, neutral, or negative. These sentiment scores were then normalized and incorporated into the overall risk evaluation. To detect bot activity, we used DBScan clustering to classify Reddit users as either bot-like or human-like based on post frequency, account age, and message similarity. This method allowed us to detect artificially inflated hype around certain tokens. The machine learning component of the project involved training an XGBoost model using historical token data labeled as rug pulls or legitimate projects. The model considered factors such as liquidity trends, volume fluctuations and market cap changes to generate a probability score predicting whether a token is a high-risk rug pull. We also included HoneyPot API to detect if a crypto coin was a honeypot based on other variables.

Challenges we ran into

We ran into issues of choosing API endpoints that were free or very cost friendly. We had issues of finding datasets to train on initially since there is not much pre-labeled data out there. We wish we had better and more data to train on for our XgBoost for better precision an accuracy. When working on the Reddit API, we had some issues with latency because of the amount of info that had to be processed. By the time we were ready to publish, we had issues making sure our project built perfectly so that we could host it on Azure.

Accomplishments that we're proud of

We are proud of creating a clean Restful API using Django that made frontend easy. We were able to find a GraphQL database for eth chain to train our model on. Crafting queries were a challenge that we eventually overcame.