Cahyono, Sugeng Dwi (2023) ADAPTIVE MOMENT ESTIMATION (ADAM) PADA DEEP LONG SHORT TERM MEMORY (DLSTM) UNTUK PERAMALAN NILAI TUKAR RUPIAH (IDR) TERHADAP MATA UANG ASING. Other thesis, UPN "Veteran" Yogyajarta.
![[thumbnail of 1. Skripsi Fulltext_123160018_Sugeng Dwi Cahyono.pdf]](http://eprints.upnyk.ac.id/style/images/fileicons/text.png) |
Text
1. Skripsi Fulltext_123160018_Sugeng Dwi Cahyono.pdf Restricted to Repository staff only Download (3MB) |
|
Text
2. Abstrak_123160018_Sugeng Dwi Cahyono.pdf Download (9kB) |
|
Text
3. Cover_123160018_Sugeng Dwi Cahyono.pdf Download (163kB) |
|
Text
4. Lembar Pengesahan_123160018_Sugeng Dwi Cahyono.pdf Download (999kB) |
|
Text
5. Daftar Isi_123160018_Sugeng Dwi Cahyono.pdf Download (141kB) |
|
Text
6. Daftar Pustaka_123160018_Sugeng Dwi Cahyono.pdf Download (83kB) |
Abstract
vii
ABSTRAK
Penelitian ini menggunakan data nilai tukar Rupiah terhadap USD dan JPY. Data yang
digunakan merupakan data nilai tukar harian dari Oktober 2014 hingga Oktober 2022. Data
tersebut telah diolah dengan melakukan normalisasi, feature sliding window, serta dilatih
dan diuji menggunakan arsitektur LSTM dengan kombinasi hyperparameter. Penelitian ini
berhasil menunjukkan bahwa penambahan hidden layer pada model LSTM dapat secara
signifikan mengurangi nilai RMSE dan MAPE, sehingga meningkatkan performa
keseluruhan model. Selain itu, hasil eksperimen menunjukkan bahwa algoritma optimizer
Adam memiliki performa training yang lebih unggul dibandingkan SGD. Khususnya, dalam
peramalan nilai tukar IDR/USD, penggunaan kombinasi stacked layer dan optimizer Adam
berhasil mengurangi MAPE sebesar 6,217%, sementara untuk nilai tukar JPY/IDR,
pengurangan MAPE mencapai 6,811%. Temuan ini menegaskan bahwa arsitektur yang
digunakan memiliki potensi untuk diimplementasikan dalam model peramalan data time
series.
Kata Kunci: LSTM, prediksi, deret waktu
viii
ABSTRACT
This research utilized the exchange rate data of Rupiah against USD and JPY. The
data consisted of daily exchange rates spanning from October 2014 to October 2022.
Preprocessing steps included normalization and feature sliding window, followed by
training and testing the data using the LSTM architecture with various hyperparameter
combinations. This study successfully demonstrated that adding hidden layers to the LSTM
model significantly reduces the values of RMSE and MAPE, thereby enhancing the overall
performance of the model. Additionally, experimental results revealed that the Adam
optimizer outperforms SGD in terms of training performance. Specifically, in forecasting
the IDR/USD exchange rate, the combination of stacked layers and the Adam optimizer led
to an 6,217% reduction in MAPE, while for the JPY/IDR exchange rate, the reduction in
MAPE reached 6,811%. These findings affirm that the employed architecture holds
potential for implementation in time series data forecasting models.
Keywords: LSTM, forecasting, timeseries
| Item Type: | Thesis (Other) |
|---|---|
| Uncontrolled Keywords: | LSTM, forecasting, timeseries |
| Subjects: | Z Bibliography. Library Science. Information Resources > ZA Information resources |
| Divisions: | Faculty of Engineering, Science and Mathematics > School of Engineering Sciences |
| Depositing User: | Eko Yuli |
| Date Deposited: | 03 Aug 2023 07:35 |
| Last Modified: | 03 Aug 2023 08:21 |
| URI: | http://eprints.upnyk.ac.id/id/eprint/36760 |
Actions (login required)
 |
View Item |