LSTM-based Price Prediction and Dynamic Risk Management in Decentralized Blockchain Protocol

This study investigates Bitcoin price prediction and dynamic risk management strategies within decentralized finance (DeFi) protocols using Long Short-Term Memory (LSTM) neural network models. The research demonstrates that the LSTM model effectively captures Bitcoin's general price trends and short-term fluctuations under typical market conditions. However, during periods of extreme volatility, the prediction model exhibits notable lag and reduced amplitude in capturing abrupt price changes, highlighting its limitations when relying solely on historical price data. Furthermore, this paper proposes a dynamic collateral ratio adjustment mechanism based on predicted price deviations, aimed at mitigating liquidation risks in DeFi lending protocols. Dynamically adjusting collateral ratios has the potential to substantially improve protocol stability compared to traditional static collateral frameworks.