Pemodelan Data Radiosonde Menggunakan Hybrid Stacking Ensemble Untuk Klasifikasi Hujan Sedang-Lebat

Hermansyah, Muhammad (2025) Pemodelan Data Radiosonde Menggunakan Hybrid Stacking Ensemble Untuk Klasifikasi Hujan Sedang-Lebat. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6025232007-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Perubahan iklim telah menyebabkan peningkatan intensitas dan frekuensi hujan lebat di wilayah tropis seperti Indonesia. Untuk mendukung sistem peringatan dini bencana hidrometeorologis, diperlukan model klasifikasi hujan yang akurat. Penelitian ini mengusulkan pendekatan Hybrid Stacking Ensemble untuk mengklasifikasikan hujan sedang–lebat berbasis data radiosonde yang selama ini belum banyak dimanfaatkan dalam sistem prediksi operasional.
Model yang dikembangkan mengintegrasikan algoritma Random Forest, XGBoost, LightGBM, dan Support Vector Machine (SVM) sebagai base learners, yang dilatih menggunakan teknik Out-of-Fold (OOF) prediction. Output probabilistik dari model-model dasar tersebut kemudian digabungkan dengan fitur asli dan digunakan sebagai input bagi meta-learner, yaitu HistGradientBoostingClassifier. Untuk mengatasi ketidakseimbangan data antara berawan-hujan ringan dan hujan sedang–lebat, digunakan teknik Synthetic Minority Over-sampling Technique (SMOTE). Dataset yang digunakan berasal dari integrasi data radiosonde dan curah hujan harian di Kota Tarakan selama periode 24 November 2015 hingga 31 Desember 2024. Evaluasi performa dilakukan menggunakan metrik precision, recall, F1-score, serta kurva precision–recall. Hasil eksperimen menunjukkan bahwa model Hybrid Stacking mencatatkan F1-score tertinggi sebesar 0.8972 dan average precision sebesar 0.955 untuk kelas hujan sedang–lebat, mengungguli semua model tunggal maupun stacking konvensional. Temuan ini menegaskan bahwa integrasi data atmosfer vertikal dengan pendekatan ensemble learning berbasis multi-algoritma dan fitur gabungan efektif dalam mendeteksi hujan sedang-lebat. Model ini memiliki potensi besar untuk diimplementasikan dalam sistem peringatan dini cuaca ekstrem secara real-time dan operasional.
================================================================================================================================
Climate change has led to an increase in the intensity and frequency of heavy rainfall in tropical regions such as Indonesia. To support early warning systems for hydrometeorological disasters, accurate rainfall classification models are essential. This study proposes a Hybrid Stacking Ensemble approach to classify moderate-to-heavy rainfall based on radiosonde data, which remains underutilized in operational forecasting systems.
The proposed model integrates Random Forest, XGBoost, LightGBM, and Support Vector Machine (SVM) as base learners, trained using Out-of-Fold (OOF) prediction. The probabilistic outputs from these base models are combined with the original features and fed into a meta-learner, namely HistGradientBoostingClassifier. To address the class imbalance between light and moderate-to-heavy rainfall events, the Synthetic Minority Over-sampling Technique (SMOTE) is employed. The dataset is derived from the integration of radiosonde observations and daily rainfall measurements in Tarakan City over the period from 24 November 2015 to 31 December 2024. Model performance is evaluated using precision, recall, F1-score, and precision–recall curve metrics. Experimental results show that the Hybrid Stacking model achieves the highest F1 score of 0.8972 and an average precision of 0.955 for the moderate-to-heavy rainfall class, outperforming all individual models and conventional stacking methods.
These findings demonstrate that integrating vertical atmospheric data with a multi-algorithm ensemble learning framework and combined feature representation is effective for detecting moderate to heavy rainfall. The proposed model holds strong potential for real-time and operational implementation in extreme weather early warning systems.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	klasifikasi hujan, hybrid stacking ensemble, radiosonde, pembelajaran mesin, data tidak seimbang
Subjects:	T Technology > T Technology (General) > T57.8 Nonlinear programming. Support vector machine. Wavelets. Hidden Markov models. T Technology > T Technology (General) > T58.62 Decision support systems
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User:	Muhammad Hermansyah
Date Deposited:	05 Aug 2025 07:43
Last Modified:	17 Sep 2025 03:42
URI:	http://repository.its.ac.id/id/eprint/126803

Actions (login required)

View Item