HAKIM, Dimara Kusuma and Gernowo, Rahmat and Nirwansyah, Anang Widhi (2026) PENGEMBANGAN MODEL PREDIKSI BANJIR DI WILAYAH PESISIR DENGAN KOMBINASI METODE JARINGAN SARAF TIRUAN DAN PENGHALUSAN EKSPONENSIAL. Doctoral thesis, UNIVERSITAS DIPONEGORO.
|
Text
cover aja.pdf Download (142kB) |
|
|
Text
00-Cover.pdf Restricted to Repository staff only Download (3MB) |
|
|
Text
01-Bab-01.pdf Download (249kB) |
|
|
Text
02-Bab-02.pdf Download (810kB) |
|
|
Text
03-Bab-03.pdf Restricted to Repository staff only Download (403kB) |
|
|
Text
04-Bab-04.pdf Restricted to Repository staff only Download (14MB) |
|
|
Text
05-Bab-05.pdf Restricted to Repository staff only Download (223kB) |
|
|
Text
06-Daftar Pustaka.pdf Download (220kB) |
|
|
Text
07-Lampiran.pdf Restricted to Repository staff only Download (16MB) |
Abstract
Banjir merupakan fenomena kompleks dengan kerugian ekonomi global mencapai USD 388,4 miliar, sehingga pengembangan model prediksi bahaya yang akurat menjadi prioritas dalam sistem manajemen bencana. Namun, efektivitas sistem saat ini terhambat oleh dua kendala teknis fundamental. Pertama, terdapat kesenjangan metode dalam penentuan bobot parameter; pendekatan berbasis pedoman standar (seperti IRBI) seringkali dianggap terlalu kaku secara nasional, sementara metode Proses Hierarki Analitik yang akomodatif terhadap karakteristik lokal memiliki ketergantungan tinggi pada persepsi pakar yang bersifat subjektif. Kedua, sistem peramalan yang ada umumnya masih berbasis pada pendekatan univariat yang hanya memproses satu variabel secara tunggal. Padahal, banjir rob dipicu oleh interaksi non-linear yang kompleks, sehingga diperlukan transisi ke pendekatan data panel (multivariat) untuk menangkap korelasi antar-variabel independen sekaligus dimensi waktu secara simultan. Untuk meminimalkan subjektivitas dalam penentuan bobot tersebut, digunakan Jaringan Saraf Tiruan (JST) yang difokuskan pada analisis kepentingan fitur. Pendekatan berbasis data ini memberikan objektivitas dalam algoritma pembobotan, di mana sistem secara otomatis mengekstraksi variabel yang berpengaruh di wilayah studi. Metodologi ini memiliki keunggulan dalam memproses hubungan non-linear yang memberikan akurasi di kisaran0,87–0,94. Untuk mengatasi limitasi pendekatan univariat, diadopsi arsitektur data panel yang mencakup data elevasi (DEMNAS), limpasan dari tutupan lahan (ESA WorldCover), jarak garis pantai, kemiringan, dan fitur cuaca historis (ERA5). Model hibrida yang menggabungkan JST dan Penghalusan Eksponensial Tripel menunjukkan performa unggul dalam memproyeksikan pola tren dan musiman pada data deret waktu. Model ini mampu menekan nilaiMAPEprediksi hingga ke kisaran3,5%–3,8%, jauh lebih presisi dibandingkan model univariate tunggal yang memilikiMAPEdi rentang11%–12%. Seluruh hasil pemodelan kemudian diimplementasikan ke dalam Sistem Informasi Geografis berbasis web untuk visualisasi bahaya banjir. Orisinalitas penelitian ini terletak pada pembobotan objektif dan transformasi arsitektur data dari univariat ke multivariat data panel sebagai salah satu tahapan mitigasi bencana.
Kata kunci:Banjir Pesisir, Jaringan Saraf Tiruan, Penghalusan Eksponensial Tripel, Pembobotan Objektif, Hibrida:Peramalan-lalu-Klasifikasi.
Flooding is a complex phenomenon with global economic losses reaching USD 388.4 billion, making the development of accurate hazard prediction models a priority in disaster management systems. However, the effectiveness of current systems is hindered by two fundamental technical constraints. First, there is a methodological gap in determining parameter weights; standard guideline-based approaches (such as IRBI) are often considered too nationally rigid, while the Analytical Hierarchy Process (AHP) method, which accommodates local characteristics, has a high dependence on subjective expert perception. Second, existing forecasting systems are generally still based on univariate approaches that process only a single variable. In reality, tidal flooding is triggered by complex non-linear interactions, necessitating a transition to a panel data (multivariate) approach to simultaneously capture correlations between independent variables and the time dimension.
To minimize subjectivity in weight determination, an Artificial Neural Network (ANN) focused on feature importance analysis is utilized. This data-driven approach provides objectivity in the weighting algorithm, where the system automatically extracts influential variables in the study area. This methodology excels in processing non-linear relationships, yielding accuracy in the range of 0.87–0.94.
To overcome the limitations of the univariate approach, a panel data architecture is adopted, encompassing elevation data (DEMNAS), runoff from land cover (ESA WorldCover), coastline distance, slope, and historical weather features (ERA5). A hybrid model combining ANN and Triple Exponential Smoothing demonstrates superior performance in projecting trends and seasonal patterns in time-series data. This model is capable of reducing the predictionMAPEvalue to the range of3.5%–3.8%, which is significantly more precise than a single univariate model that has aMAPEin the range of11%–12%. All modeling results are then implemented into a web-based Geographic Information System for flood hazard visualization. The originality of this research lies in the objective weighting and the transformation of data architecture from univariate to multivariate panel data as a critical stage in disaster mitigation.
Keywords:Coastal Flooding, Artificial Neural Network, Triple Exponential Smoothing, Objective Weighting, Hybrid:Forecast-then-Classify
| Item Type: | Thesis (Doctoral) |
|---|---|
| Uncontrolled Keywords: | Banjir Pesisir, Jaringan Saraf Tiruan, Penghalusan Eksponensial Tripel, Pembobotan Objektif, Hibrida:Peramalan-lalu-Klasifikasi |
| Subjects: | Sciences and Mathemathic |
| Divisions: | Postgraduate Program > Doctor Program in Information System |
| Depositing User: | ekana listianawati |
| Date Deposited: | 07 Sep 2026 08:06 |
| Last Modified: | 07 Sep 2026 08:06 |
| URI: | https://eprints2.undip.ac.id/id/eprint/60618 |
Actions (login required)
![]() |
View Item |
