Perbandingan Model Attention-LSTM dan Transformer dalam Peramalan Curah Hujan sebagai Mitigasi Risiko Bencana Banjir Provinsi Jawa Timur

Yudiawati, Marcella Prastika (2026) Perbandingan Model Attention-LSTM dan Transformer dalam Peramalan Curah Hujan sebagai Mitigasi Risiko Bencana Banjir Provinsi Jawa Timur. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5006221021-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Perubahan iklim telah meningkatkan variabilitas dan intensitas curah hujan di Provinsi Jawa Timur yang berdampak pada meningkatnya risiko bencana hidrometeorologi seperti banjir dan tanah longsor. Oleh karena itu, diperlukan metode peramalan yang akurat untuk mendukung perencanaan dan mitigasi risiko. Penelitian ini bertujuan untuk membandingkan kinerja model Attention Long Short-Term Memory (Attention-LSTM) dan Transformer dalam memprediksi curah hujan di Jawa Timur dengan fokus pada tiga stasiun wilayah, yaitu Pasuruan, Malang, dan Banyuwangi menggunakan data curah hujan harian BMKG periode 2021-2025. Model Attention-LSTM memanfaatkan mekanisme attention untuk memberikan bobot pada setiap time step sehingga mampu mengekstraksi informasi penting dari keluaran LSTM, sedangkan model Transformer menggunakan mekanisme self-attention untuk menangkap ketergantungan jangka panjang secara paralel tanpa bergantung pada struktur berurutan. Data curah hujan diproses melalui tahap pembersihan, normalisasi, dan pembagian data menjadi data training dan testing. Kedua model dilatih kemudian kinerja model dievaluasi menggunakan metrik Root Mean Squared Error (RMSE) dan Mean Absolute Error (MAE). Hasil penelitian menunjukkan bahwa kombinasi hyperparameter terbaik berbeda pada setiap wilayah, yang mengindikasikan bahwa karakteristik data curah hujan memengaruhi konfigurasi model yang optimal. Berdasarkan hasil evaluasi dan visualisasi prediksi, Attention-LSTM menghasilkan performa yang lebih konsisten serta mampu merepresentasikan pola curah hujan aktual dengan lebih baik dibandingkan Transformer, khususnya pada data dengan karakteristik fluktuatif dan jumlah data yang relatif terbatas. Oleh karena itu, Attention-LSTM dipilih sebagai model terbaik dalam penelitian ini. Temuan penelitian ini diharapkan dapat menjadi bahan pertimbangan dalam pengembangan model prediksi curah hujan berbasis deep learning sebagai bagian dari sistem peringatan dini (early warning system) guna mendukung pengambilan keputusan dalam mitigasi risiko bencana hidrometeorologi di Provinsi Jawa Timur.
=====================================================================================================================================
Climate change has increased the variability and intensity of rainfall in East Java Province, leading to a higher risk of hydrometeorological disasters such as floods and landslides. Therefore, accurate rainfall forecasting methods are essential to support disaster planning and risk mitigation. This study aims to compare the performance of the Attention Long Short-Term Memory (Attention-LSTM) and Transformer models for rainfall forecasting in East Java, focusing on three meteorological stations located in Pasuruan, Malang, and Banyuwangi using daily rainfall data obtained from the Indonesian Agency for Meteorology, Climatology, and Geophysics (BMKG) for the period of 2021–2025. The Attention-LSTM model employs an attention mechanism to assign weights to each time step, enabling the extraction of important information from LSTM outputs, while the Transformer model utilizes a self-attention mechanism to capture long-term dependencies in parallel without relying on sequential processing. The rainfall data were processed through data cleaning, normalization, and splitting into training and testing sets. Both models were trained, and their performance was evaluated using Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). The results indicate that the optimal hyperparameter combinations differ across the study areas, suggesting that rainfall data characteristics influence the optimal model configuration. Based on the evaluation results and prediction visualization, the Attention-LSTM model demonstrated more consistent performance and was better able to represent actual rainfall patterns than the Transformer model, particularly for highly fluctuating rainfall data and relatively limited datasets. Therefore, the Attention-LSTM model was selected as the best-performing model in this study. These findings are expected to contribute to the development of deep learning-based rainfall forecasting models as part of an Early Warning System (EWS) to support decision-making in hydrometeorological disaster risk mitigation in East Java Province.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Attention-LSTM, Curah hujan, Deep learning, Peramalan, Transformer, Attention-LSTM, Deep learning, Forecasting, Rainfall, Transformer
Subjects:	Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry) Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) Q Science > QC Physics > QC866.5 Climatology--Forecasting. Q Science > QC Physics > QC925 Rain and rainfall
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Actuaria > 94203-(S1) Undergraduate Thesis
Depositing User:	Marcella Prastika Yudiawati
Date Deposited:	16 Jul 2026 03:39
Last Modified:	16 Jul 2026 03:39
URI:	http://repository.its.ac.id/id/eprint/135125

Actions (login required)

View Item