Pemodelan Nilai Ekspor Indonesia Menggunakan Metode Hybrid ARIMAX-FFNN dan Extreme Gradient Boosting (XGBoost)

Arya, Andra Citta Passadhi (2023) Pemodelan Nilai Ekspor Indonesia Menggunakan Metode Hybrid ARIMAX-FFNN dan Extreme Gradient Boosting (XGBoost). Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6003211035-Master_Thesis.pdf - Accepted Version Restricted to Registered users only until 1 September 2025. Download (2MB) | Request a copy

Abstract

Peramalan ekspor Indonesia yang akurat memiliki urgensi penting sebagai salah satu acuan untuk merumuskan target pertumbuhan ekonomi nasional. Nilai ekspor Indonesia diduga dipengaruhi oleh faktor-faktor eksternal di antaranya kurs rupiah dan nilai impor. Oleh karena itu, perlu untuk memodelkan peramalan nilai ekspor Indonesia dengan melibatkan faktor-faktor tersebut. Metode peramalan yang digunakan adalah peramalan hybrid yang menggabungkan dua model yaitu model linier dan nonlinier sehingga melalui metode hybrid diharapkan dapat menghasilkan ramalan yang lebih akurat. Tahap pertama dilakukan pemodelan linier yaitu dengan model Autoregressive Integrated Moving Average with Exogenous Variabel (ARIMAX). Selanjutnya, di tahap kedua dilakukan pemodelan hybrid yang bekerja dengan menggunakan input residual dari model linier dengan pendekatan machine learning yaitu Feed Forward Neural Network (FFNN) untuk menangkap pola nonlinier. Selain pemodelan hybrid, dilakukan juga pemodelan dengan menggunakan algoritma maching learning lainnya yaitu Extreme Gradient Boosting (XGBoost). Tujuan penelitian ini adalah membandingkan akurasi model peramalan hybrid ARIMAX-FFNN dan XGBoost. Data yang digunakan adalah data ekspor Indonesia bulanan periode 1 Januari 2009 – 31 Desember 2022 yang diambil dari data sekunder Badan Pusat Statistik (BPS). Pemilihan model terbaik berdasarkan nilai Mean Absolute Percentage Error (MAPE) yang terkecil. Hasilnya, XGBoost memiliki performa peramalan yang lebih baik dibandingkan ARIMAX-FFNN karena memiliki nilai MAPE yang lebih kecil yaitu sebesar 6,60 % sedangkan ARIMAX-FFNN memiliki nilai MAPE sebesar 7,52%.
===================================================================================================================================
Accurate forecasting of Indonesian exports has an important urgency as a reference for formulating national economic growth targets. Indonesia's export value is thought to be influenced by external factors including the rupiah exchange rate and import values. Therefore, it is necessary to model forecasting the value of Indonesian exports by involving these factors. The forecasting method used is hybrid forecasting which combines two models, namely linear and nonlinear models so that through the hybrid method it is expected to produce more accurate forecasts. The first stage is linear modeling, namely the Autoregressive Integrated Moving Average with Exogenous Variable (ARIMAX) model. Furthermore, in the second stage, hybrid modeling is carried out which works by using the residual input from a linear model with a machine learning approach, namely Feed Forward Neural Network (FFNN) to capture nonlinear patterns. In addition to hybrid modeling, modeling is also carried out using another maching learning algorithm, namely Extreme Gradient Boosting (XGBoost). The purpose of this study is to compare the accuracy of the hybrid ARIMAX-FFNN and XGBoost forecasting models. The data used is monthly Indonesian export data for the period 1 January 2009 – 31 December 2022 taken from secondary data from the Statistics Indonesia (BPS). Selection of the best model is based on the smallest Mean Absolute Percentage Error (MAPE). As a result, XGBoost has better forecasting performance than ARIMAX-FFNN because it has a smaller MAPE value of 6,60% while ARIMAX-FFNN has a MAPE value of 7,52%.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Ekspor, Peramalan; Hybrid, Arimax-FFNN, XGBoost
Subjects:	Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry) Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49101-(S2) Master Thesis
Depositing User:	Andra Citta Passadhi Arya
Date Deposited:	11 Aug 2023 02:03
Last Modified:	11 Aug 2023 02:03
URI:	http://repository.its.ac.id/id/eprint/104432

Actions (login required)

View Item