FBN.2024.10
Bitcoin Price Prediction
Machine learning engineer, sole build
TensorFlow, Keras, LSTM, Pandas, 2024
LSTM
recurrent memory for sequences
2
error measures: MSE and RMSE
BTC
one brutally noisy series
The problem
Cryptocurrency prices are noisy, non-stationary and merciless to naive models.
The approach
An LSTM with a careful preprocessing pipeline of cleaning, scaling and windowing, measured with MSE and RMSE and backtested visually against real prices.
The outcome
A forecaster with its error stated plainly, and a hard-won intuition for financial time series.
What it does
3 capabilitiesPreprocessing done properly
Cleaning, scaling and windowing: the unglamorous work that decides a model's fate.
Validation without flattery
Performance stated in MSE and RMSE, not in cherry-picked charts.
Backtests you can see
Predicted and actual curves overlaid, showing exactly where the model holds and where it breaks.
Next exhibit — XI
EatFit!