Perbandingan Metode ARIMA dan Backpropagation Neural Network dalam Peramalan Harga Ethereum (ETH)

Abstract

The fluctuating nature of Ethereum prices necessitates forecasting methods that can produce accurate predictions to support investment decision-making. This study aims to compare the performance of the Autoregressive Integrated Moving Average (ARIMA) and Backpropagation Neural Network methods for forecasting Ethereum prices and determine which method yields better predictive performance. The study uses historical Ethereum price data, which are processed through data preprocessing, model development, and testing stages. The ARIMA model is developed based on the characteristics of time series data, while the Backpropagation Neural Network model is designed to capture nonlinear patterns in the dataset. The performance of both models is evaluated using Mean Absolute Percentage Error (MAPE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). The results indicate that the Backpropagation Neural Network achieves a MAPE of 2.793%, lower than the ARIMA model's 3.942%. Based on these results, the Backpropagation model achieves a forecasting accuracy of 97.21%, while the ARIMA model achieves 96.06%. Although the ARIMA model yields slightly lower MSE and RMSE, the MAPE-based evaluation indicates that the Backpropagation Neural Network provides superior forecasting performance for Ethereum price data in this study. Therefore, the Backpropagation Neural Network is a more effective alternative than ARIMA for forecasting Ethereum prices using the dataset employed in this research.