Implementasi Hybrid TabNet-XGBoost Berbasis Data Radiosonde untuk Prediksi Intensitas Hujan di Kota Surabaya

Puryani, Annabel Gracia (2026) Implementasi Hybrid TabNet-XGBoost Berbasis Data Radiosonde untuk Prediksi Intensitas Hujan di Kota Surabaya. Undergraduate thesis, UPN Veteran Jawa Timur.

	Text (22083010048-cover) cover.pdf Download (502kB)
	Text (22083010048-bab1) bab 1.pdf Download (142kB)
	Text (22083010048-bab2) bab 2.pdf Restricted to Repository staff only until 14 September 2028. Download (608kB) \| Request a copy
	Text (22083010048-bab3) bab 3.pdf Restricted to Repository staff only until 14 September 2028. Download (435kB) \| Request a copy
	Text (22083010048-bab4) bab 4_compressed.pdf Restricted to Repository staff only until 14 September 2028. Download (874kB) \| Request a copy
	Text (22083010048-bab5) bab 5.pdf Download (93kB)
	Text (22083010048-daftarpustaka) dapus.pdf Download (72kB)
	Text (22083010048-lampiran) lampiran.pdf Restricted to Repository staff only Download (262kB) \| Request a copy

Abstract

Rainfall is one of the weather elements that plays an important role in community activities and is closely related to the risk of hydrometeorological disasters, especially flooding in urban areas such as the city of Surabaya. The high variability in rainfall intensity and the availability of extreme data pose challenges in developing a prediction model capable of generating low prediction errors, while the benefits of using rainosonde data in a machine learning approach combining TabNet and regression models still need to be evaluated. This study aims to develop a rainfall intensity prediction model for the city of Surabaya based on rainosonde data from 2019 to 2025 through variable selection using Pearson correlation and TabNet, lag determination using the Cross-Correlation Function (CCF), as well as the implementation of three scenarios: Hybrid TabNet–XGBoost, Hybrid TabNet–XGBoost with extreme-data handling, and Hybrid TabNet–SVR. Extreme-data handling was performed by excluding rainfall data exceeding 80 mm during the training process, while model parameter optimization was conducted using Optuna and evaluation was performed using RMSE, MAE, and MAPE. The research results show that Hybrid TabNet–XGBoost yields an RMSE of 19.1404 mm, an MAE of 11.8515 mm, and a MAPE of 18.53%, whereas Hybrid TabNet–XGBoost with extreme-data handling yields an RMSE of 14.4482 mm, an MAE of 10.1127 mm, and a MAPE of 19.03%; Meanwhile, Hybrid TabNet–SVR yields an RMSE of 20.8319 mm, an MAE of 10.6938 mm, and a MAPE of 12.36%. Therefore, Hybrid TabNet–XGBoost with extreme-data handling provides the best performance based on RMSE and MAE, whereas Hybrid TabNet–SVR yields a lower MAPE. The innovation of this study lies in the integration of TabNet as a variable selection method with XGBoost and SVR, the use of CCF-based lags, as well as testing the impact of handling extreme rainfall data across three model scenarios. The third model is subsequently used to generate rainfall intensity predictions for the year 2026, which are presented through a Streamlit-based application in the form of tables and visualizations.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Damaliana, Aviolla Terza	NIDN002089402	aviolla.terza.sada@upnjatim.ac.id
Thesis advisor	Pratama, Alfan Rizaldy	NUPTK938777678130112	alfan.fasilkom@upnjatim.ac.id

Subjects:

H Social Sciences > HA Statistics

Divisions:

Faculty of Computer Science > Departemen of Data Science

Depositing User:

Annabel Gracia Puryani

Date Deposited:

14 Sep 2026 08:02

Last Modified:

14 Sep 2026 08:17

URI:

https://repository.upnjatim.ac.id/id/eprint/59984