Prasetyo, Daniel Bergas (2025) Prediksi Harga Emas Menggunakan Model Bi-GRU Dengan Monte Carlo Dropout Berdasarkan Data Makroekonomi. Undergraduate thesis, UPN Veteran Jawa Timur.
![[img]](https://repository.upnjatim.ac.id/style/images/fileicons/text.png) |
Text (cover)
21081010167.-cover.pdf Download (8MB) |
|
Text (bab 1)
21081010167.-bab1.pdf Download (295kB) |
|
Text (bab 2)
21081010167.-bab2.pdf Restricted to Repository staff only until 19 December 2027. Download (727kB) |
|
Text (bab 3)
21081010167.-bab3.pdf Restricted to Repository staff only until 19 December 2027. Download (508kB) |
|
Text
21081010167.-bab4.pdf Restricted to Repository staff only until 19 December 2027. Download (6MB) |
|
Text
21081010167.-bab5.pdf Download (289kB) |
|
Text (daftar pustaka)
21081010167.-daftarpustaka.pdf Download (179kB) |
|
Text (lampiran)
21081010167.-lampiran.pdf Restricted to Repository staff only Download (233kB) |
Abstract
This study evaluates the performance of a Bidirectional Gated Recurrent Unit (BI-GRU) model combined with Monte Carlo Dropout for forecasting gold prices (XAU) using macroeconomic indicators such as CPI, DXY, S&P 500, and crude oil prices over the period from May 6, 2015, to May 1, 2025. The research is motivated by the need for predictive tools capable of capturing non-linear relationships while quantifying uncertainty in highly volatile market conditions. The primary objective is to assess the performance of the BI-GRU model with Monte Carlo Dropout and compare it with the standard BI-GRU model. The research procedures include collecting daily and monthly secondary data, performing data preprocessing (merging, cleaning, and Min-Max normalization), determining the time-series window, tuning hyperparameters (data ratio, window size, number of GRU units, and dropout rate), training the model using the Adam optimizer, and evaluating performance using MAE, RMSE, and R² metrics. Uncertainty estimation and prediction intervals are generated through Monte Carlo sampling. The results reveal a strong correlation between gold prices and the S&P 500 index (r ≈ 0.93) as well as CPI (r ≈ 0.89). The best configuration—70:30 data ratio, a window size of 90, 128 GRU units, and a dropout rate of 0.3—achieved an MAE of 0.0479, an RMSE of 0.0587, and an R² of approximately 0.8845. In comparison, the standard BI-GRU model without Monte Carlo Dropout recorded an MAE of 0.0454, an RMSE of 0.0586, and an R² of 0.8751, indicating lower accuracy and weaker generalization capability. Based on the multi-iteration forecasting results, the model demonstrates strong predictive performance up to the 60th iteration, while predictions beyond the 90th and 120th iterations show reduced quality and diminished sensitivity to changing market trends.
| Item Type: | Thesis (Undergraduate) | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Contributors: |
|
||||||||||||
| Subjects: | Q Science > QA Mathematics > QA76.87 Neural computers | ||||||||||||
| Divisions: | Faculty of Computer Science > Departemen of Informatics | ||||||||||||
| Depositing User: | Bergas Daniel Prasetyo | ||||||||||||
| Date Deposited: | 19 Dec 2025 09:17 | ||||||||||||
| Last Modified: | 19 Dec 2025 09:17 | ||||||||||||
| URI: | https://repository.upnjatim.ac.id/id/eprint/48420 |
Actions (login required)
 |
View Item |