IPB University Logo

SCIENTIFIC REPOSITORY

IPB University Scientific Repository collects, disseminates, and provides persistent and reliable access to the research and scholarship of faculty, staff, and students at IPB University

AI Repository
 
Building and Categories


      View Item 
      •   IPB Repository
      • Final Assignments
      • Master Final Assignments
      • MF - School of Data Science, Mathematic and Informatics
      • View Item
      •   IPB Repository
      • Final Assignments
      • Master Final Assignments
      • MF - School of Data Science, Mathematic and Informatics
      • View Item
      JavaScript is disabled for your browser. Some features of this site may not work without it.

      Analisis Perbandingan Model NARNN, SARIMA-NARNN, dan Holt-Winters-NARNN dalam Prediksi Nilai Ekspor Indonesia

      Thumbnail
      View/Open
      Cover (496.1Kb)
      Fulltext (1.142Mb)
      Lampiran (253.6Kb)
      Date
      2026
      Author
      Akbar, Raihan
      Mangku, I Wayan
      Silalahi, Bib Paruhum
      Metadata
      Show full item record
      Abstract
      Nilai ekspor adalah total pemasukan suatu negara dari penjualan barang dan jasa ke luar negeri, biasanya diukur dalam periode tertentu menggunakan mata uang internasional (USD). Indikator ini mencerminkan kondisi pasar dan ekonomi sebuah negara: semakin banyak negara lain bergantung padanya, semakin kuat pula stabilitas ekonominya. Nilai ekspor Indonesia terdiri dari ekspor nonmigas menyumbang sekitar 248.83 miliar USD sedangkan ekspor migas tercatat sebesar 15.88 miliar USD di sepanjang 2021. Sektor non-migas terdiri dari sektor pertanian, hasil hutan, hasil ternak, kerajinan tangan, industri dan produk mineral hasil tambang. Selain itu, dilansir dari Badan Pusat Statistik mencatat bahwa nilai ekspor Indonesia pada Oktober 2024 mencapai 24.41 miliar USD meningkat sekitar 10% dibandingkan bulan Oktober tahun sebelumnya. Peningkatan nilai ekspor yang terjadi di Indonesia dapat menjadi potensi yang cukup kuat untuk bersaing dalam kompetisi pasar global. Meski secara kuantitas masih jauh dibawah negara-negara eksportir utama dunia, namun penting untuk menjaga tren pertumbuhan positif yang berkontribusi signifikan terhadap stabilitas ekonomi nasional. Penelitian terkait nilai ekspor yang membantu perencanaan strategi ekspor adalah prediksi nilai ekspor melalui data historisnya. Berdasarkan latar belakang tersebut, penelitian tesis ini bertujuan untuk (1) Menganalisis karakteristik dari nilai ekspor Indonesia melalui analisis deskriptif dan dekomposisi data guna mengidentifikasi komponen tren, musiman dan residual. (2) Membandingkan kinerja model SARIMA-NARNN, HW-NARNN dan NARNN dalam menghasilkan prediksi terbaik terhadap nilai ekspor Indonesia. Penelitian ini menggunakan data sekunder nilai ekspor Indonesia yang bersumber dari publikasi resmi Badan Pusat Statistik berupa nilai Free on Board bulanan per komoditas yang mencakup harga barang beserta biaya transportasi hingga ke pelabuhan. Data diintegrasikan melalui Microsoft Excel untuk periode Januari 2013 hingga Juni 2025 dalam satuan Juta US$, kemudian dimodelkan menggunakan RStudio dan MATLAB 2024a. Prosedur penelitian dibagi menjadi lima tahap sistematis. Tahap pertama, pra-pemrosesan data, meliputi pengumpulan data, analisis deskriptif melalui plotting untuk mengidentifikasi tren, musiman, dan pencilan, serta pembagian data menjadi 90% data latih (Januari 2013–Maret 2024) dan 10% data uji (April 2024–Juni 2025). Tahap kedua memodelkan komponen non-linear menggunakan Nonlinear Autoregressive Neural Network (NARNN), diawali normalisasi Min-Max Scaling ke rentang [-1,1], penentuan arsitektur jaringan (jumlah feedback delay dan hidden layer) secara trial and error, dan pelatihan untuk meminimalkan MSE. Tahap ketiga adalah pemodelan komponen linear menggunakan dua pendekatan: Seasonal Autoregressive Integrated Moving Average (SARIMA) melalui transformasi Box-Cox, uji stasioneritas ADF beserta differencing bila diperlukan, identifikasi orde dari plot ACF/PACF, seleksi model berdasarkan AIC terkecil, dan uji diagnostik residual serta Holt-Winters Additive (HWA) dan Holt-Winters Multiplicative (HWM) dengan parameter pemulusan optimal (a,ß,?). Dari kedua model klasik terbaik diekstraksi residual yang diasumsikan mengandung informasi non-linear yang belum tertangkap. Tahap keempat merekonstruksi model hybrid dengan mendenormalisasi prediksi residual NARNN, memprediksi residual data uji, lalu menjumlahkannya kembali dengan prediksi linear untuk membentuk hybrid SARIMA-NARNN dan hybrid HW-NARNN. Tahap terakhir mengevaluasi kinerja pada data uji menggunakan Root Mean Square Error, Mean Absolute Error, dan Mean Absolute Percentage Error, serta membandingkan ketiga model: NARNN, hybrid SARIMA-NARNN, dan hybrid HW-NARNN. Hasil dari penelitian ini adalah hibridisasi meningkatkan model HW tetapi tidak meningkatkan SARIMA, dengan HWA–NARNN sebagai model hybrid terbaik. Namun, kinerja keseluruhan terbaik dicapai oleh NARNN(12,3), dengan RMSE sebesar 1279.07, MAE sebesar 997.41, dan MAPE sebesar 4.46%. Temuan ini menunjukkan bahwa nilai ekspor Indonesia mengandung dinamika nonlinear yang substansial dan bahwa NARNN merupakan pendekatan prediksi paling efektif di antara model yang dipertimbangkan.
       
      Export value is the total revenue a country earns from selling goods and services abroad, typically measured over a given period in an international currency (USD). This indicator reflects a country's market and economic conditions: the more other countries depend on it, the stronger its economic stability. Indonesia's export value comprises non-oil-and-gas exports, contributing about USD 248.83 billion, and oil-and-gas exports, recorded at USD 15.88 billion, throughout 2021. The non-oil-and-gas sector spans agriculture, forestry products, livestock products, handicrafts, industry, and mineral mining products. In addition, Badan Pusat Statistik reported that Indonesia's export value in October 2024 reached USD 24.41 billion, an increase of roughly 10% compared with October of the previous year. This rising export value represents considerable potential for Indonesia to compete in the global market. Although its volume remains far below that of the world's leading exporters, maintaining a positive growth trend is important, as it contributes significantly to national economic stability. One line of research on export value that supports export-strategy planning is the prediction of export value from its historical data. Against this background, this thesis aims to (1) analyze the characteristics of Indonesia's export value through descriptive analysis and data decomposition in order to identify trend, seasonal, and residual components; and (2) compare the performance of the SARIMA-NARNN, HW-NARNN, and NARNN models in producing the best predictions of Indonesia's export value. This study uses secondary data on Indonesia's export value sourced from official publications of Statistics Indonesia, in the form of monthly Free on Board (FOB) values per commodity, which include the price of goods together with transportation costs to the port. The data were integrated using Microsoft Excel for the period January 2013 to June 2025, expressed in millions of US$, and subsequently modeled using RStudio and MATLAB 2024a. The research procedure is divided into five systematic stages. The first stage, data pre-processing, comprises data collection, descriptive analysis through plotting to identify trend, seasonality, and outliers, and splitting the dataset into 90% training data (January 2013-March 2024) and 10% testing data (April 2024-June 2025). The second stage models this nonlinear component using a Nonlinear Autoregressive Neural Network (NARNN), beginning with Min-Max Scaling normalization to the range [-1,1], determination of the network architecture (the number of feedback delays and hidden layers) by trial and error, and training to minimize the MSE. The third stage is linear-component modeling using two approaches: Seasonal Autoregressive Integrated Moving Average (SARIMA) via Box-Cox transformation, the ADF stationarity test with differencing where required, order identification from ACF/PACF plots, model selection based on the lowest AIC, and residual diagnostic testing, as well as Holt-Winters Additive (HWA) and Holt-Winters Multiplicative (HWM) with optimal smoothing parameters (a,ß,?). From the two best classical models, residuals are extracted, assumed to contain nonlinear information not yet captured. The fourth stage reconstructs the hybrid model by denormalizing the NARNN residual predictions, predicting the residuals of the testing data, and then adding them back to the linear predictions to form the hybrid SARIMA-NARNN and hybrid HW-NARNN. The final stage evaluates performance on the testing data using the Root Mean Square Error, Mean Absolute Error, and Mean Absolute Percentage Error, and compares the three models: NARNN, hybrid SARIMA-NARNN, and hybrid HW-NARNN. The results show that hybridization improves the HW model but not SARIMA, with HWA–NARNN as the best hybrid model. However, the best overall performance is achieved by NARNN(12,3), with an RMSE of 1279.07, an MAE of 997.41, and a MAPE of 4.46%. These findings indicate that Indonesia's export value contains substantial nonlinear dynamics and that NARNN is the most effective predictive approach among the models considered.
       
      URI
      http://repository.ipb.ac.id/handle/123456789/176990
      Collections
      • MF - School of Data Science, Mathematic and Informatics [138]

      Copyright © 2020 Library of IPB University
      All rights reserved
      Contact Us | Send Feedback
      Indonesia DSpace Group 
      IPB University Scientific Repository
      UIN Syarif Hidayatullah Institutional Repository
      Universitas Jember Digital Repository
        

       

      Browse

      All of IPB RepositoryCollectionsBy Issue DateAuthorsTitlesSubjectsThis CollectionBy Issue DateAuthorsTitlesSubjects

      My Account

      Login

      Application

      google store

      Copyright © 2020 Library of IPB University
      All rights reserved
      Contact Us | Send Feedback
      Indonesia DSpace Group 
      IPB University Scientific Repository
      UIN Syarif Hidayatullah Institutional Repository
      Universitas Jember Digital Repository