Mushoddaq, Mohammad Zaki (2026) Prediction Commercial Chicken Egg Prices in Sidoarjo Regency Using Gated Recurrent Unit (GRU)-TPE. Undergraduate thesis, UPN Veteran Jawa Timur.
|
Text (Cover)
19081010017.-Cover.pdf Download (803kB) | Preview |
|
|
Text (Bab 1)
19081010017.-bab1.pdf Download (136kB) | Preview |
|
![[img]](https://repository.upnjatim.ac.id/style/images/fileicons/text.png) |
Text (Bab 2)
19081010017.-bab2.pdf Restricted to Repository staff only until 7 July 2029. Download (286kB) |
|
|
Text (Bab 3)
19081010017.-bab3 .pdf Restricted to Repository staff only until 7 July 2029. Download (281kB) |
|
|
Text (Bab 4)
19081010017.-bab4.pdf Restricted to Repository staff only until 7 July 2029. Download (977kB) |
|
|
Text (Bab 5)
19081010017.-bab5.pdf Download (120kB) | Preview |
|
|
Text (Daftar pustaka)
19081010017.-daftarpustaka.pdf Download (124kB) | Preview |
|
|
Text (Lampiran)
19081010017.-lampiran.pdf Restricted to Repository staff only Download (69kB) |
Abstract
Commercial chicken eggs are a strategic food commodity with highly volatile prices, reaching up to 82% difference between the lowest and highest values based on daily SIMPONI Ternak data from Sidoarjo Regency for the 2020–2025 period. This price instability significantly affects farmers, traders, and regional food security, underscoring the need for an accurate prediction system accessible to non-technical users. This study aimed to evaluate the performance of a Gated Recurrent Unit (GRU) model optimized with the Tree-structured Parzen Estimator (TPE) via Optuna, assess the contribution of calendar-based exogenous features to prediction accuracy, and deploy the best model as a Streamlit-based web application. A quantitative experimental approach was employed with three input scenarios: univariate (price only), bivariate (price + is_holiday), and trivariate (price + is_holiday + dtoh_norm). A total of 2,192 daily observations were chronologically split at a 70:15:15 ratio, with each scenario independently optimized over 100 trials. The GRU-TPE-Tri model achieved the best performance on the test data, with a MAPE of 0.6064%, MAE of Rp164.80, and RMSE of Rp359.16, outperforming the unoptimized LSTM baseline by 16.35% in MAPE. Adding the dtoh_norm feature reduced the MAE by 28.03% compared to the bivariate scenario, demonstrating that a quantitative representation of temporal distance to holidays is more informative than a binary signal alone. The best model was successfully deployed as a functional web application prototype without programming expertise. Subsequent research may add other exogenous variables, perform multi-step prediction, or use hybrid methods.
| Item Type: | Thesis (Undergraduate) | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Contributors: |
|
||||||||||||
| Subjects: | T Technology > T Technology (General) T Technology > T Technology (General) > T385 Computer Graphics T Technology > T Technology (General) > T55.4-60.8 Industrial engineering. Management engineering T Technology > T Technology (General) > T58.6-58.62 Management Information Systems |
||||||||||||
| Divisions: | Faculty of Computer Science > Departemen of Informatics | ||||||||||||
| Depositing User: | Mohammad Zaki Mushoddaq | ||||||||||||
| Date Deposited: | 07 Jul 2026 04:17 | ||||||||||||
| Last Modified: | 07 Jul 2026 04:17 | ||||||||||||
| URI: | https://repository.upnjatim.ac.id/id/eprint/54693 |
Actions (login required)
 |
View Item |