Akmal, Muhammad Azkiya (2026) IMPLEMENTASI TEMPORAL FUSION TRANSFORMER (TFT) UNTUK PREDIKSI PENJUALAN JANGKA PENDEK PAKET DATA TELKOMSEL DI JAWA TIMUR. Undergraduate thesis, UPN Veteran Jawa Timur.
|
Text (Cover)
Cover.pdf Download (1MB) | Preview |
|
|
Text (Bab 1)
Bab 1.pdf Download (248kB) | Preview |
|
![[img]](https://repository.upnjatim.ac.id/style/images/fileicons/text.png) |
Text (Bab 2)
Bab 2.pdf Restricted to Repository staff only until 6 May 2028. Download (725kB) |
|
|
Text (bab 3)
Bab 3.pdf Restricted to Repository staff only until 6 May 2028. Download (558kB) |
|
|
Text (Bab 4)
Bab 4.pdf Restricted to Repository staff only until 6 May 2028. Download (1MB) |
|
|
Text (Bab 5)
Bab 5.pdf Restricted to Repository staff only until 6 May 2028. Download (223kB) |
|
|
Text (Daftar Pustaka)
Daftar Pustaka.pdf Download (204kB) | Preview |
|
|
Text (Lampiran)
Lampiran.pdf Restricted to Repository staff only until 6 May 2028. Download (376kB) |
Abstract
Accurate short-term sales forecasting for data packages is crucial for telecommunications providers to optimize inventory availability and operational efficiency. However, the complexity of transaction patterns, influenced by temporal fluctuations and regional demographic conditions, poses significant challenges in modeling. This research implements the Temporal Fusion Transformer (TFT) architecture to predict short-term sales of Telkomsel data packages across 15 city clusters in East Java. The innovation proposed in this study is the integration of the Seasonal-Trend Decomposition using Loess (STL) method during the preprocessing stage. The decomposed trend, seasonal, and residual components are utilized as observed past inputs to enrich the model's information, alongside static variables (cluster and package type) and dynamic variables (hour, day, and national holidays). The model is evaluated using a quantile regression approach to generate prediction intervals. Results indicate superior model performance, achieving a Mean Absolute Error (MAE) of 3.69 and a q-Risk (P50) of 0.12. Variable Selection Network (VSN) analysis reveals that Cluster (14.21%) and STL Trend (8.45%) are the most dominant factors influencing sales, while the attention mechanism validates strong daily (24-hour) and weekly (7-day) cyclical patterns. As a practical contribution, this predictive model is implemented into a web-based interactive dashboard system to support management in real-time strategic decision-making.
| Item Type: | Thesis (Undergraduate) | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Contributors: |
|
||||||||||||
| Subjects: | Q Science > QA Mathematics > QA76.6 Computer Programming Q Science > QA Mathematics > QA76.87 Neural computers |
||||||||||||
| Divisions: | Faculty of Computer Science > Departemen of Data Science | ||||||||||||
| Depositing User: | Muhammad Azkiya Akmal | ||||||||||||
| Date Deposited: | 13 May 2026 06:56 | ||||||||||||
| Last Modified: | 13 May 2026 08:13 | ||||||||||||
| URI: | https://repository.upnjatim.ac.id/id/eprint/51611 |
Actions (login required)
 |
View Item |