Crypto Price Prediction

About this project

This project explores whether market information can be combined with an external signal to build a richer Bitcoin forecasting pipeline. Historical BTC prices are collected through Yahoo Finance, while Bitcoin Wikipedia revision history is retrieved through the MediaWiki API and the revision comments are classified using a pretrained Hugging Face sentiment model.

The data is aligned by date and transformed into sliding time-series windows, using the previous 60 days to predict the following day's price. A three-layer LSTM network with dropout is trained using the Adam optimizer and mean squared error loss, with the resulting predictions evaluated against unseen market data.

The project brings together data collection, sentiment analysis, time-series preprocessing, recurrent neural networks, and model evaluation in one pipeline, exploring how traditional numerical signals and external textual information can be incorporated into the same forecasting problem.