<![CDATA[Highly volatile price changes in crypto assets like Bitcoin complicate precise forecasting. The Extreme Learning Machine (ELM) artificial neural network approach offers high computational speed but is prone to performance instability due to random weight initialization and manual hyperparameter determination. To overcome this problem, this study proposes a hybrid PSO-ELM algorithm to automate the search for optimal parameters, namely the number of hidden neurons and the regression regularization penalty. Evaluated using a five-fold Walk-Forward Validation to prevent data leakage, the model comparatively tested five input feature scenarios based on technical indicators, which are mathematical calculations from historical prices to identify market patterns. Results demonstrate the hybrid PSO-ELM outperforms conventional static models, reducing average error (MAPE) by 18.67 percent. The cross-scenario comparison reveals that applying the Simple Moving Average technical indicator yields the best forecasting model, achieving a 26.65 percent error reduction and a final MAPE accuracy of 2.15 percent. The contribution of this research is providing empirical evidence that automatic parameter optimization combined with a random fluctuation filtering feature is proven to be more robust and accurate in responding to extreme volatility compared to the stacking of various complex derivative momentum indicators.]]>
Copyrights © 2026
