<![CDATA[Bitcoin is a crypto asset for investment. It can give high profit, but it also has high risk because the price changes very fast and is not stable. To reduce the risk of loss, we need a prediction system that can read price changes well. This research aims to model and predict the closing price of Bitcoin using network activity data (on-chain metrics). The method used is Deep Learning with the BiLSTM algorithm. This method is chosen because it can process data in two directions (forward and backward), so it can learn patterns better than standard LSTM. The dataset is taken from the public Blockchain network using BigQuery, from August 18, 2011, to February 6, 2026, with 5,287 daily data. The model uses the main input active_spending_addresses and two volatility indicators: Percent of Top Range (PTR) and Percent Low Range (PLR). Before modeling, the data is processed using a sliding window of 60 days, with 90% training data and 10% testing data. The results show that the BiLSTM model has high accuracy, with MAE 2.958, RMSE 3.905, and MAPE 3.22%. The comparison shows that BiLSTM is better than other models. LSTM has MAPE 29.06%, and MLP has MAPE 4.01%. In conclusion, BiLSTM can handle extreme crypto market changes very well, so it gives stable and accurate Bitcoin price predictions.]]>
Copyrights © 2026
