Model Prediksi Harga Bitcoin Menggunakan Algoritma SBI-GRU dan SBI-LSTM

Abstract

The main issue in this study is the high volatility of bitcoin prices. Based on our observations, we found that bitcoin price volatility ranges from 35% to 85%, making bitcoin investments quite risky. Therefore, the purpose of this study is to create a bitcoin price prediction model that can follow this volatility pattern. We created a neural network (NN)-based Bitcoin price prediction model using the Stacked-Bidirectional Gated Recurrent Unit (SBI-GRU) and Stacked-Bidirectional Long Short-Term Memory (SBI-LSTM) algorithms. Both algorithms are suitable and have been proven capable of predicting Bitcoin prices. Furthermore, we also used hyperparameter tuning to improve the accuracy of this prediction model. The tuning parameters used were epoch, batch size, and optimizer. Our research results show that this prediction model is capable of predicting bitcoin prices and their volatility patterns. The benchmarks are as follows: the SBI-GRU model has a MAPE value of 0.0254 and a correlation coefficient of 0.9975. Meanwhile, the SBI-LSTM model has a MAPE value of 0.0333 and a correlation coefficient of 0.9963. The interpretation of these results is that both prediction models, using the SBI-GRU and SBI-LSTM methods, are able to predict bitcoin prices very well. A MAPE value of less than 0.05 makes the prediction model accurate, while a correlation coefficient of 0.99 means that the model is able to follow the volatility pattern.