Prediksi Harga USOIL Berdasarkan Data Historis Menggunakan Metode Stacked LSTM

Crude oil (USOIL) serves as a strategic commodity vital to the global
economy, yet it exhibits extremely high price volatility. These sharp fluctuations,
driven by geopolitical factors, OPEC+ policies, and market sentiment, create
significant uncertainty. Traditional prediction methods often fail to capture the
non-linear dynamics and long-term dependencies inherent in oil price data. This
study aims to develop and implement a deep learning model to enhance the
accuracy of USOIL price prediction.
The method employed in this research is Stacked Long Short-Term
Memory (Stacked LSTM), designed to learn complex temporal patterns
hierarchically. This study utilizes daily historical data of USOIL prices (Open,
High, Low, Close, and Volume). The data underwent pre-processing stages,
including normalization via Min-Max Scaler and transformation using a 30-day
sliding window technique. To evaluate model reliability, this study applies
cross-validation using the TimeSeriesSplit method.
The results indicate that the Stacked LSTM model demonstrates excellent
performance in predicting USOIL prices. Quantitatively, the model achieved an
R-squared (R²) sebesar 0.95, demonstrating its ability to explain 95% of the
price variability. Other evaluation metrics also exhibited high precision, with a
Root Mean Squared Error (RMSE) of 0.0175 and a Mean Absolute Error
(MAE) of 0.0127. The primary conclusion suggests that the Stacked LSTM
architecture is highly effective for short-term prediction (1-day), although its
performance declines over longer time horizons.