ML-001

Machine Learning Models

Defi Data has collected +50 million of rows of twitter frequency and sentiment data across the top 400 tokens. This equates to about 9 full months of data at minute resolution. This dataset will be used to run ML experiments and build models to correlate price with twitter volume and sentiment metrics.

Research has shown around 70% prediction accuracy for bitcoin. However bitcoin is highly liquiditywith many institutional players. We aim to look at lesser known tokens in less liquid markets to find out if twitter has a larger effect in these markets. Our hypothesis is that these mascent tokens are largely driven by online culture & small community.

While most projects focus on a single market we will aim to train across multiple markets in that aim that a model can potentially expres not only tweets to price but consider type of market (volume, market cap, volatility etc).