Analisis Peramalan Nilai Impor Nonmigas Indonesia Menggunakan Pendekatan Deep Learning Dengan Model Stacked-Long Short-Term Memory dan Stacked-Gated Recurrent Unit

Gaya, Joycelin Gracelda Resi (2026) Analisis Peramalan Nilai Impor Nonmigas Indonesia Menggunakan Pendekatan Deep Learning Dengan Model Stacked-Long Short-Term Memory dan Stacked-Gated Recurrent Unit. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221121-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Nilai impor nonmigas merupakan salah satu indikator penting dalam perdagangan internasional Indonesia yang mendukung kebutuhan bahan baku, barang modal, dan barang konsumsi. Peramalan nilai impor nonmigas diperlukan untuk mendukung perencanaan ekonomi dan pengambilan kebijakan yang lebih tepat. Namun, karakteristik data yang berfluktuasi dan bersifat non-linear menyebabkan metode deret waktu konvensional sering mengalami keterbatasan dalam menangkap pola temporal jangka panjang. Oleh karena itu, penelitian ini membandingkan kinerja model deep learning Stacked-Long Short-Term Memory (Stacked-LSTM) dan Stacked-Gated Recurrent Unit (Stacked-GRU) dalam meramalkan nilai impor nonmigas Indonesia guna memperoleh model yang paling optimal. Data yang digunakan merupakan data bulanan periode Januari 1993 hingga Desember 2025 yang bersumber dari Badan Pusat Statistik (BPS). Penelitian dilakukan melalui serangkaian tahapan yang meliputi preparasi data, penentuan parameter optimal, pelatihan model, evaluasi performa menggunakan Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Squared Error (MSE), dan Mean Absolute Percentage Error (MAPE) dengan MAPE sebagai evaluasi metrik utama, serta forecasting. Hasil penelitian menunjukkan bahwa model terbaik pada kedua arsitektur menggunakan konfigurasi 2 layer dengan 25 dan 12 unit, dropout rate 0,2 dan 0,1, batch size 16, serta learning rate 0,005. Model Stacked-LSTM menghasilkan nilai MAPE sebesar 9,36% pada data training dan 6,83% pada data testing, lebih rendah dibandingkan model Stacked-GRU yang menghasilkan MAPE sebesar 9,62% dan 7,14%. Selain itu, visualisasi hasil prediksi menunjukkan bahwa Stacked-LSTM lebih mampu mengikuti pola dan fluktuasi data aktual dibandingkan Stacked-GRU, begitupun juga dengan hasil forecasting menunjukkan kedua model memiliki kecenderungan penurunan nilai pada periode prediksi, namun Stacked-LSTM menghasilkan pola peramalan yang lebih selaras dengan data historis. Dengan demikian, Stacked-LSTM terbukti lebih optimal digunakan untuk peramalan nilai impor nonmigas Indonesia.
=====================================================================================================================================
Because it reflects the demand for raw materials, capital goods, and consumer goods. Forecasting non-oil and gas imports is important to support economic planning and policy making. However, the fluctuating and non-linear nature of import data makes it difficult for conventional time series methods to capture complex long-term patterns. Therefore, this study compares the performance of Stacked Long Short-Term Memory (Stacked-LSTM) and Stacked Gated Recurrent Unit (Stacked-GRU) models to determine the most suitable model for forecasting Indonesia’s non-oil and gas import value. This study uses monthly non-oil and gas import data from January 1993 to December 2025 obtained from Statistics Indonesia (BPS). The research process includes data preparation, model development, hyperparameter tuning, model training, performance evaluation using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Squared Error (MSE), and Mean Absolute Percentage Error (MAPE), and forecasting. Among these metrics, MAPE was used as the primary performance evaluation metric to determine the best model. The results show that the best model configuration for both architectures consists of two layers with 25 and 12 units, dropout rates of 0.2 and 0.1, a batch size of 16, and a learning rate of 0.005. The Stacked-LSTM model achieved MAPE values of 9.36% on the training data and 6.83% on the testing data, while the Stacked-GRU model achieved MAPE values of 9.62% and 7.14%, respectively. In addition, the prediction plots show that Stacked-LSTM is better at following the patterns and fluctuations of the actual data. The forecasting results indicate that both models predict a downward trend in non-oil and gas import values, but Stacked-LSTM produces forecasts that are more consistent with historical data patterns. Therefore, Stacked-LSTM is considered the better model for forecasting Indonesia’s non-oil and gas import value.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Deep Learning, Nilai Impor Nonmigas, Peramalan, Stacked-LSTM, Stacked-GRU, Deep Learning, Non-Oil and Gas Imports, Forecasting, Stacked-LSTM, Stacked-GRU
Subjects:	H Social Sciences > HA Statistics > HA30.3 Time-series analysis
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Joycelin Gracelda Resi Gaya
Date Deposited:	30 Jul 2026 01:47
Last Modified:	30 Jul 2026 01:47
URI:	http://repository.its.ac.id/id/eprint/140026

Actions (login required)

View Item