The N-HiTS (Neural Hierarchical Interpolation for Time Series) model is a deep learning framework designed for time series forecasting. It decomposes input data into hierarchical levels, each capturing different temporal patterns. Through an interpolation mechanism, the model generates intermediate forecasts, which are recursively refined for accuracy. This approach allows N-HiTS to effectively capture both short-term fluctuations and long-term trends.
In this study, I used the N-HiTS model to predict the price of Bitcoin for the next 30 days using Onchain data from the past 180 days. The modeling and training were conducted using the PyTorch, PyTorch Lightning, and PyTorch Forecasting libraries.
Figure A shows the predicted and actual prices after the training process for the validation data, while Figure B shows the forecast for the next 30 days.The training data includes 376 features taken from the cryptoquant platform.
Written by CryptoOnchain