Analisis Kinerja XGBoost Menggunakan Bayesian Optimization dalam Prediksi Harga Ethereum

Abstract

Cryptocurrency is a digital innovation in the financial sector that has revolutionized the global transaction system through blockchain technology. One of the main challenges in the crypto domain today is determining the price of cryptocurrencies, which are highly volatile. Ethereum, one of the largest cryptocurrencies, exhibits complex volatility patterns that require a robust predictive system. This study aims to compare the performance of the standard XGBoost algorithm with XGBoost optimized using Bayesian Optimization in predicting daily Ethereum prices based on time series data from 2016 to June 2025. The dataset includes price-related features such as open, high, low, volume, and percentage price change. The modeling process consists of several stages including feature engineering, time series-based data splitting, and model training. Model performance was evaluated using three primary metrics: MAE, RMSE, and R² Score. The evaluation results show that the standard XGBoost model achieved an MAE of 80.8926 (3.12%), RMSE of 114.1457 (4.40%), and an R² Score of 0.9723. Meanwhile, the optimized model using Bayesian Optimization achieved an MAE of 70.7241 (2.73%), RMSE of 102.5334 (3.96%), and an R² Score of 0.9777. These results indicate that Bayesian Optimization helps improve the model's prediction accuracy. This study concludes that the XGBoost model with a Bayesian optimization approach yields superior and more effective performance in forecasting Ethereum prices based on time series data.