Azharudin, Nabilah (2025) Peramalan Harga Minyak Mentah Dunia Jenis Brent Menggunakan Metode Long Short-Term Memory (LSTM) dan Bidirectional Long Short-Term Memory (BiLSTM). Other thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 5003211093-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
5003211093-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (11MB) | Request a copy |
Abstract
Minyak mentah Brent merupakan salah satu acuan utama dalam perdagangan minyak global yang pergerakan harganya sangat berpengaruh terhadap stabilitas ekonomi dunia. Oleh karena itu, meramalkan harga minyak Brent secara akurat sangat penting, terutama untuk pengambilan keputusan strategis di sektor energi dan ekonomi makro. Penelitian ini bertujuan untuk meramalkan harga minyak mentah Brent menggunakan dua metode deep learning, yaitu Long Short-Term Memory (LSTM) dan Bidirectional Long Short-Term Memory (BiLSTM), sebagai metode analisis deret waktu nonlinear. Pemodelan LSTM dan BiLSTM dilakukan dengan menerapkan metode Grid Search untuk tuning hyperparameter, seperti jumlah neuron, tingkat dropout, jumlah epoch, ukuran batch, dan kombinasi input lag berdasarkan hasil analisis fungsi autokorelasi parsial (PACF). Model LSTM terbaik memiliki konfigurasi 128 neuron, tingkat dropout 0,2, epoch 50, batch size 64, dan input lag (Xt-2, Xt-4, Xt-5, Xt-6), dengan hasil evaluasi menunjukkan MAPE sebesar 2,59%, MAE 2,0437, dan RMSE 2,6074. Sementara itu, model BiLSTM dikembangkan untuk menangkap dependensi temporal dua arah (forward dan backward) guna meningkatkan akurasi peramalan. Konfigurasi terbaik model BiLSTM mencakup 64 neuron, tingkat dropout 0,3, epoch 50, batch size 64, dan input lag (Xt-2, Xt-4, Xt-6) dengan MAPE 1,82%, MAE 1,4469, dan RMSE 1,8835. Hasil peramalan menunjukkan bahwa LSTM mampu mengenali tren historis, sedangkan BiLSTM lebih adaptif terhadap fluktuasi dan dinamika kompleks data harga.
=========================================================================================================================================
Brent crude oil is a major benchmark in global oil trade, with its price movements significantly impacting global economic stability. Therefore, accurately forecasting Brent oil prices is crucial, particularly for strategic decision-making in the energy and macroeconomic sectors. This research aims to forecast Brent crude oil prices using two deep learning methods, Long Short-Term Memory (LSTM) and Bidirectional Long Short-Term Memory (BiLSTM), as nonlinear time series analysis. LSTM and BiLSTM modeling are conducted by applying the Grid Search method to tune hyperparameters, such as the number of neurons, dropout rate, number of epochs, batch size, and input lag combinations based on the results of the partial autocorrelation function (PACF) analysis. The best LSTM model has a configuration of 128 neurons, dropout rate of 0,2, 50 epochs, batch size of 64, and input lags at (Xt-2, Xt-4, Xt-5, Xt-6) with evaluation results showing a MAPE of 2,59%, MAE of 2,0437, and RMSE of 2,6074. Meanwhile, the BiLSTM model was developed to capture two-way temporal dependencies (forward and backward) to improve forecasting accuracy. The best configuration of BiLSTM model includes 64 neurons, dropout rate of 0,3, 50 epochs, batch size of 64, and input lags at (Xt-2, Xt-4, Xt-6) with a MAPE of 1,82%, MAE of 1,4469, and RMSE of 1,8835. The forecasting results show that LSTM is capable to recognize historical trends, while BiLSTM is more adaptive to fluctuations and complex dynamics of data price.
Actions (login required)
 |
View Item |