Bitcoin, as the most representative cryptocurrency, has an extremely volatile price, and traditional financial theories are difficult to fully explain its market behavior. The high volatility and complexity of Bitcoin’s price pose a great challenge to investment decisions. This study proposes a bitcoin exchange rate prediction method based on a combined ARIMA-LSTM model, which improves the prediction accuracy by combining traditional time series analysis with deep learning techniques. Methodologically, an LSTM neural network is first constructed to capture the nonlinear characteristics of the bitcoin price, then an ARIMA model is built to analyze the linear trend, and finally the prediction results of the two models are optimally combined by using the CRITIC weight assignment method. The experiment uses the bitcoin closing price data from September 1, 2021 to December 31, 2024 for validation. The results show that the combined ARIMA-LSTM model significantly outperforms the single model in terms of forecasting performance, with a mean absolute error (MAE) of 0.0002, a root mean square error (RMSE) of 0.0003, and a mean absolute percentage error (MAPE) of 0.0006, which are 0.0071, 0.004, and 0.0051 lower than that of the ARIMA model, respectively. Empirical analysis shows that the combined model can more accurately capture the changing law of bitcoin price by integrating the advantages of linear and nonlinear prediction methods, which provides effective technical support for digital currency investment.
