Advanced Bitcoin Price Prediction: Real-Time Analysis using Flask, LSTM and Boosting Algorithms

The explosive growth and subsequent collapse of cryptocurrencies has opened a world of possibilities never seen before, led by one of the top digital currencies in existence today – Bitcoin. Nothing less than the global penetration and adoption of Bitcoin around the world has, despite high liquidity, that level of extreme volatility–something which threatens to put investors and institutions in harm’s way making resilient forecasting methods a necessity. Linear statistical models (e.g., ARIMA and GARCH etc.) are not good enough to capture the non-linear relationship in financial market, therefore we observe machine learning (ML) and deep learning (DL) algorithms for financial time series. In this paper, we suggest an ensemble prediction method integrating LSTM-based models and boosting methods (XGBoost, LightGBM, CatBoost) in order to retain both precision and interpretability. Historical Bitcoin trading volumes, moving averages, and sentiment indicators were collected during the model prepping stage. Results showed that the LSTM and boosting could complement each other in exploring sequence dependency and stabilizing non-linear feature learning. The ensemble approach with a combination of this two-folds strategy had an improved performance in terms of the reduction of the bias and variance simultaneously. We then developed a web application with Flask to deploy these models and perform real-time prediction and plotting on the Bitcoin price movement.