<![CDATA[Predicting cryptocurrency is difficult because it has high volatility, where prices can experience spikes or declines due to market dynamics. This study focuses on NameCoin, one of the oldest altcoins originating from Bitcoin. NameCoin was selected because it has relatively stable and extensive historical data. The objective of this study is to evaluate the performance of the Bidirectional Recurrent Neural Network (BiRNN) in predicting NameCoin price movements. This study employs an experimental method using historical data as input for the training process. Hyperparameter tuning is conducted systematically using four different scenarios to obtain the optimal model configuration. The dataset is divided into 80% for training the model and 20% for testing the performance of the trained model. Model performance is evaluated using RMSE, MSE, MAPE, coefficient of determination (R²), Directional Statistic (D-Stat), and loss value as indicators of model accuracy and stability. The experimental results show that Scenario 1 produces the most optimal performance, with RMSE = 0.0216, MAPE = 2.59%, R² = 0.9899, D-Stat = 53.71%, and the smallest loss value of 0.0012. These performance metrics indicate that the BiRNN model effectively captures nonlinear trends and accurately predicts the direction of price movements. Conversely, Scenario 3 had the worst performance, with a MAPE of 10.19%. By comparing these scenarios, it is clear that the configuration in Scenario 1 outperforms the others in terms of prediction accuracy and model stability against data fluctuations.]]>
Copyrights © 2025
