Comparative Analysis of LSTM and XGBoost for Cryptocurrency Trend Prediction

Abstract

It is difficult to predict cryptocurrency market trends due to volatility and extraneous factors. In this study, the XGBoost and LSTM models are contrasted in terms of forecasting bullish and bearish trends in the Ethereum market. Technical indicators like Simple Moving Average (SMA), Exponential Moving Average (EMA), and trading volume were employed as features based on past market data. Market tendencies were established through changes in closing price on a day-to-day basis, and training and testing the two models applied was carried out on normalized data under supervised learning. Results in experiments show that XGBoost exhibited superior performance at 84.58%, which was entirely above the 50.45% achieved using LSTM. The differential in performance implicates superior performance by XGBoost in establishing nonlinear relations among input features and market tendencies. The results show that XGBoost is more suitable for market trend classification, while LSTM cannot generalize and overfit time-series patterns. The results offer possibilities for sequential model enhancement by employing more technical indicators or hybrid architectures that combine time-series analysis and ensemble learning techniques.