Peramalan Volume Transaksi Uang Elektronik Menggunakan SARIMA-LSTM dengan Integrasi Headline Berita Berbasis IndoBERT dan Google Trends

Wardana, Nabila Sya'bani (2026) Peramalan Volume Transaksi Uang Elektronik Menggunakan SARIMA-LSTM dengan Integrasi Headline Berita Berbasis IndoBERT dan Google Trends. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6026241014-Master_Thesis.pdf Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Pertumbuhan transaksi uang elektronik di Indonesia memerlukan proyeksi volume jangka pendek bagi otoritas sistem pembayaran dan penyedia jasa. Model konvensional umumnya belum menangkap pola linier, musiman, dan non-linier secara bersamaan, serta belum memanfaatkan sinyal perilaku digital publik. Penelitian ini mengusulkan model hibrida SARIMA-LSTM untuk meramalkan volume transaksi uang elektronik bulanan di Indonesia, dengan SARIMA mengestimasi struktur linier dan musiman serta LSTM memodelkan residual untuk menangkap pola non-linier. Sebagai variabel eksogen, digunakan proksi arah pemberitaan media yang diklasifikasikan menjadi kategori pendorong, penghambat, dan informatif menggunakan IndoBERT (F1-Macro 91,10%), serta indeks penelusuran Google Trends. Evaluasi dilakukan dengan skema rolling origin pada horizon satu hingga tiga bulan menggunakan metrik MAE, RMSE, dan MAPE. Pada data uji, model hibrida SARIMA-LSTM konsisten menghasilkan kesalahan lebih rendah dibanding SARIMA maupun LSTM tunggal di seluruh horizon, meskipun dengan selisih yang moderat. Kontribusi variabel eksogen bersifat terbatas dan bergantung horizon, hanya proporsi berita informatif yang secara konsisten menurunkan kesalahan di seluruh horizon dengan selisih kecil, sedangkan proporsi pendorong, penghambat, dan skor neto justru memperburuk; dari sisi Google Trends, penurunan kesalahan terlihat pada horizon dua dan tiga bulan, sementara pada horizon satu bulan kontribusinya belum konsisten
===========================================================================================================================================
The growth of electronic money transactions in Indonesia requires short-term volume projections for payment system authorities and service providers. Conventional models generally have not captured linear, seasonal, and non-linear patterns simultaneously, nor have they exploited signals of public digital behavior. This study proposes a hybrid SARIMA-LSTM model to forecast the monthly volume of electronic money transactions in Indonesia, in which SARIMA estimates the linear and seasonal structure while LSTM models the residuals to capture non-linear patterns. As exogenous variables, it employs a proxy for the direction of media coverage—classified into driving, inhibiting, and informative categories using IndoBERT (F1-Macro of 91.10%)—together with the Google Trends search index. Evaluation is conducted using a rolling-origin scheme over one- to three-month horizons with the MAE, RMSE, and MAPE metrics. On the test data, the hybrid SARIMA-LSTM consistently produces lower errors than either the standalone SARIMA or LSTM across all horizons, albeit by a moderate margin. The contribution of the exogenous variables is limited and horizon-dependent: only the proportion of informative news consistently reduces the error across all horizons, by a small margin, whereas the proportions of driving and inhibiting news and the net score instead worsen it; for Google Trends, error reductions appear at the two- and three-month horizons, while at the one-month horizon its contribution is not consistent.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Google Trends, Hibrida SARIMA-LSTM, IndoBERT, Peramalan Deret Waktu, Uang Elektronik, Electronic Money, Google Trends, Hybrid SARIMA-LSTM,IndoBERT, Time Series Forecasting
Subjects:	H Social Sciences > HB Economic Theory > Economic forecasting--Mathematical models.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 59101-(S2) Master Thesis
Depositing User:	Nabila Sya'bani Wardana
Date Deposited:	28 Jul 2026 01:46
Last Modified:	28 Jul 2026 01:46
URI:	http://repository.its.ac.id/id/eprint/138209

Actions (login required)

View Item