Peramalan Harga Minyak Mentah Dunia dengan Menggunakan Gated Recurrent Unit (GRU)

Sadewa, Naufal (2024) Peramalan Harga Minyak Mentah Dunia dengan Menggunakan Gated Recurrent Unit (GRU). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5026201136-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (3MB) | Request a copy

Abstract

Minyak mentah memegang peranan penting dalam perekonomian global, mempengaruhi aktivitas konsumsi dan produksi di berbagai negara. Fluktuasi harga minyak mentah berdampak besar terhadap stabilitas ekonomi, termasuk di Indonesia. Ketidakstabilan harga minyak dapat menyebabkan kesulitan dalam mengelola kebutuhan domestik dan menyusun Anggaran Pendapatan dan Belanja Negara (APBN). Oleh karena itu, peramalan harga minyak mentah sangat penting untuk membantu pemerintah mengambil keputusan yang tepat dalam menanggapi perubahan pasar.
Tugas Akhir ini menggunakan metode Gated Recurrent Unit (GRU), salah satu jenis Recurrent Neural Network (RNN) yang efektif untuk memodelkan hubungan temporal dalam data time series. GRU diterapkan untuk meramalkan harga minyak mentah WTI dengan menggunakan pendekatan univariat dan multivariat. Hasil penelitian menunjukkan bahwa model GRU terbaik adalah model dengan pembagian data 60%:40% dan pendekatan univariat, yang menghasilkan nilai Mean Absolute Percentage Error (MAPE) sebesar 1,75%. Parameter optimal yang digunakan meliputi 16 batch size, 5 epochs, dan 128 units dengan sequence length 1.
Model GRU menunjukkan akurasi yang lebih baik dibandingkan model Radial Basis Function Neural Network (RBFNN) sebesar 36,44%, Extreme Learning Machine (ELM) sebesar 13,09%, dan Long-Short Term Memory (LSTM) sebesar 0,79%, meskipun membutuhkan waktu eksekusi yang lebih lama dari RBFNN dan ELM. Waktu eksekusi GRU tercatat 2624,88% lebih lama dibanding RBFNN dan 180345,22% lebih lama dibanding ELM, sementara tercatat 40,01% lebih cepat dibanding LSTM. Hasil peramalan juga menunjukkan bahwa proporsi data dan normalisasi mempengaruhi akurasi dan kecepatan eksekusi model. Ringkasan hasil menunjukkan bahwa GRU memberikan hasil peramalan yang konsisten dan akurat dengan Mean Absolute Percentage Error (MAPE) rendah pada kedua jenis data, univariat dan multivariat. Namun, peramalan multivariat tidak selalu menghasilkan akurasi yang lebih baik dibandingkan univariat, menunjukkan bahwa variabilitas data dan pemilihan parameter model memiliki pengaruh signifikan terhadap akurasi dan performa model peramalan.
========================================================================================================================
Crude oil plays a crucial role in the global economy, affecting consumption and production activities in various countries. Fluctuations in crude oil prices significantly impact economic stability, including in Indonesia. Oil price instability can lead to difficulties in managing domestic needs and drafting the State Budget (APBN). Therefore, forecasting crude oil prices is essential to help the government make informed decisions in responding to market changes.
This undergraduate thesis employs the Gated Recurrent Unit (GRU) method, a type of Recurrent Neural Network (RNN) that is effective in modeling temporal relationships in time series data. GRU is applied to forecast WTI crude oil prices using univariate and multivariate approaches. The research results indicate that the best GRU model is the one with a 60%:40% data split and a univariate approach, achieving a Mean Absolute Percentage Error (MAPE) of 1,75%. The optimal parameters used include a batch size of 16, 5 epochs, and 128 units with a sequence length of 1.
The GRU model demonstrates better accuracy compared to the Radial Basis Function Neural Network (RBFNN) by 36,44%, the Extreme Learning Machine (ELM) by 13,09%, and the Long-Short Term Memory (LSTM) by 0,79%, despite requiring longer execution time than RBFNN and ELM. GRU's execution time is recorded as 2624,88% longer than RBFNN and 180345,22% longer than ELM, while being 40,01% faster than LSTM. Forecasting results also indicate that data proportions and normalization affect the accuracy and execution speed of the model. The summary of results shows that GRU provides consistent and accurate forecasting results with low Mean Absolute Percentage Error (MAPE) on both univariate and multivariate data. However, multivariate forecasting does not always yield better accuracy than univariate, indicating that data variability and model parameter selection significantly influence the accuracy and performance of forecasting models.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	forecasting, crude oil, time series, recurrent neural network, gated recurrent unit, radial basis function neural network, extreme learning machine, long-short term memory, peramalan, minyak mentah, time series, recurrent neural network, gated recurrent unit, radial basis function neural network, extreme learning machine, long-short term memory.
Subjects:	H Social Sciences > HB Economic Theory > Economic forecasting--Mathematical models. Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 57201-(S1) Undergraduate Thesis
Depositing User:	Naufal Sadewa
Date Deposited:	01 Aug 2024 05:50
Last Modified:	01 Aug 2024 05:51
URI:	http://repository.its.ac.id/id/eprint/111280

Actions (login required)

View Item