Perbandingan Kinerja Metode LSTM dan XGBoost dengan Optimasi Bayesian dalam Prediksi Curah Hujan Berbasis Indeks ENSO dan IOD

Mas'adah, Aliyah Nur (2026) Perbandingan Kinerja Metode LSTM dan XGBoost dengan Optimasi Bayesian dalam Prediksi Curah Hujan Berbasis Indeks ENSO dan IOD. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5006221077-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (10MB) | Request a copy

Abstract

Curah hujan di Indonesia, khususnya wilayah Surabaya, memiliki variabilitas tinggi yang dipengaruhi oleh fenomena iklim global seperti El Niño Southern Oscillation (ENSO) dan Indian Ocean Dipole (IOD). Ketidakpastian hidrometeorologi ini menimbulkan risiko non-tradisional yang signifikan pada sektor agrikultur, kebencanaan, dan finansial. Penelitian ini bertujuan untuk membandingkan kinerja metode Long Short-Term Memory (LSTM) dan Extreme Gradient Boosting (XGBoost) yang dioptimasi menggunakan bayesian optimization berbasis Tree-structured Parzen Estimator (TPE) untuk memprediksi curah hujan bulanan. Pemodelan dilakukan melalui dua skenario, yaitu tanpa musiman dan dengan musiman menggunakan transformasi siklik (month sin dan month cos). Hasil penelitian menunjukkan bahwa penambahan fitur musiman secara signifikan meningkatkan akurasi dan mempercepat konvergensi model. Berdasarkan hasil pengujian pada data testing, metode XGBoost terbukti lebih superior dibandingkan LSTM. Model terbaik diraih oleh model 1 XGBoost dengan skenario musiman (Curah hujan + Musiman) dengan nilai RMSE sebesar 4,1098, MAE sebesar 2,4128, dan SMAPE sebesar 61,66%. Sebaliknya, arsitektur LSTM terbaik diperoleh pada model 4 skenario musiman (Curah Hujan + Niño 3.4 + DMI + Musiman) dengan nilai RMSE sebesar 4.2738, MAE sebesar 2.5840, dan SMAPE sebesar 70.46%. Hasil peramalan rekursif 20 periode ke depan menggunakan model terbaik secara logis mampu merekonstruksi pola iklim musiman lokal. Implikasi penelitian ini dapat mendukung perencanaan tata kota, memperkuat mitigasi risiko banjir, dan mengembangkan catastrophe modeling untuk analisis risiko aktuaria di industri asuransi.
=====================================================================================================================================
Rainfall in Indonesia, particularly in the Surabaya region, exhibits high variability influenced by global climate phenomena such as the El Niño Southern Oscillation (ENSO) and the Indian Ocean Dipole (IOD). This hydrometeorological uncertainty poses significant non-traditional risks to the agricultural, disaster management, and financial sectors. This study aims to compare the performance of Long Short-Term Memory (LSTM) and Extreme Gradient Boosting (XGBoost) methods optimized using bayesian optimization based on the Tree-structured Parzen Estimator (TPE) to predict monthly rainfall. The modeling was conducted through two scenarios: without seasonality and with seasonality using cyclic transformations (month sin and month cos). The results indicate that incorporating seasonal features significantly enhances accuracy and accelerates model convergence. Based on the evaluation of the testing data, the XGBoost method proved to be superior to LSTM. The best overall model was achieved by XGBoost model 1 with the seasonal scenario ( Rainfall + Seasonal) variable and temporal components), yielding an RMSE of 4.1098, a MAE of 2.4128, and a SMAPE of 61.66%. Conversely, the best LSTM architecture was obtained in seasonal model 4 ( Rainfall + Niño 3.4 + DMI + Seasonal), resulting in an RMSE of 4.2738, a MAE of 2.5840, dan a SMAPE of 70.46%. Recursive forecasting for the next 20 periods using the superior model logically reconstructed the local seasonal climate patterns. The implications of this research can support urban planning, strengthen flood risk mitigation, and develop catastrophe modeling for actuarial risk analysis in the insurance industry.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Bayesian Optimization, Curah Hujan, ENSO, IOD, LSTM, XGBoost,Bayesian Optimization, ENSO, IOD, LSTM, Rainfall, XGBoost.
Subjects:	Q Science Q Science > Q Science (General) Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) Q Science > QA Mathematics > QA76.9.D343 Data mining. Querying (Computer science)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Actuaria > 94203-(S1) Undergraduate Thesis
Depositing User:	Aliyah Nur Mas`adah
Date Deposited:	17 Jul 2026 04:05
Last Modified:	17 Jul 2026 04:05
URI:	http://repository.its.ac.id/id/eprint/135281

Actions (login required)

View Item