Prediksi Kebakaran Hutan Menggunakan Pengembangan Model Regresi Hibrida Berbasis Data Satelit

Vianney, Benedictus Augusta (2026) Prediksi Kebakaran Hutan Menggunakan Pengembangan Model Regresi Hibrida Berbasis Data Satelit. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6026232012-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Kebakaran hutan merupakan bencana lingkungan yang besar sehingga mampu menyebabkan ancaman seperti kerusakan ekosistem, hilangnya keanekaragaman hayati, dan kerusakan ekonomi. Berbagai macam faktor seperti perubahan iklim maupun aktivitas manusia mempengaruhi peningkatan jumlah kasus dan tingkat keparahan kebakaran hutan. Metode prediksi tradisional seringkali memiliki keterbatasan dalam memperoleh dan ketepatan informasi terkait kebakaran hutan. Maka dari itu, pengembangan metode prediksi kebakaran hutan menjadi penting untuk upaya penanggulangan bencana kebakaran hutan yang akan terjadi. Penelitian ini mengusulkan pemanfaatan data satelit MODIS untuk membangun model prediksi dengan menawarkan pendekatan yang lebih menyeluruh yang diintegrasikan dengan berbagai variabel lain seperti data cuaca, informasi vegetasi, dan informasi kedekatan titik api dengan bangunan atau jalan. Penelitian ini mengembangkan model regresi hibrida berbasis data satelit guna meningkatkan akurasi prediksi kebakaran hutan pada wilayah California, Georgia, Texas, dan Florida. Data utama diperoleh dari instrumen MODIS yang kemudian diintegrasikan dengan data cuaca harian, vegetasi (NDVI), serta informasi antropogenik berupa kedekatan titik api dengan bangunan dan jalan. Tahapan pra-pemrosesan meliputi seleksi lokasi menggunakan reverse geocoding, penanganan data hilang, serta rekayasa fitur untuk menghasilkan representasi variabel yang lebih informatif, termasuk indikator intensitas kebakaran dan risiko infrastruktur. Model prediksi dibangun menggunakan pendekatan stacking yang menggabungkan kemampuan Random Forest, Gradient Boosting, dan Support Vector Regression sebagai base learners, sementara meta-learner menggunakan regresi Ridge. Seluruh model dioptimalkan melalui hyperparameter tuning berbasis Genetic Algorithm untuk memastikan stabilitas dan performa terbaik. Evaluasi menggunakan metrik R², RMSE, dan MAE menunjukkan bahwa model hibrida secara konsisten melampaui performa model tunggal, terutama dalam hal generalisasi terhadap data baru. Analisis feature importance mengidentifikasi bahwa variabel termal satelit, kondisi meteorologis, dan faktor kedekatan dengan infrastruktur memiliki pengaruh terbesar terhadap prediksi. Temuan ini menegaskan bahwa integrasi data multivariabel berbasis satelit dan lingkungan memberikan landasan yang lebih komprehensif untuk mendukung sistem peringatan dini kebakaran hutan serta strategi mitigasi berbasis data.
================================================================================================================================
Forest fires are a major environmental disaster that can cause threats such as ecosystem damage, loss of biodiversity, and economic damage. Various factors such as climate change and human activities contribute to an increase in the number of cases and severity of forest fires. Traditional prediction methods often have limitations in obtaining accurate information related to forest fires. Therefore, the development of forest fire prediction methods is important for efforts to mitigate future forest fire disasters. This study proposes the use of MODIS satellite data to build a prediction model by offering a more comprehensive approach that is integrated with various other variables such as weather data, vegetation information, and information on the proximity of fire points to buildings or roads. This study developed a hybrid regression model based on satellite data to improve the accuracy of forest fire predictions in California, Georgia, Texas, and Florida. The main data was obtained from MODIS instruments, which were then integrated with daily weather data, vegetation (NDVI), and anthropogenic information in the form of the proximity of fire points to buildings and roads. The pre-processing stages include location selection using reverse geocoding, handling missing data, and feature engineering to produce more informative variable representations, including fire intensity and infrastructure risk indicators. The prediction model was built using a stacking approach that combines the capabilities of Random Forest, Gradient Boosting, and Support Vector Regression as base learners, while the meta-learner uses Ridge regression. All models are optimized through Genetic Algorithm-based hyperparameter tuning to ensure stability and optimal performance. Evaluation using R², RMSE, and MAE metrics shows that the hybrid model consistently outperforms single models, especially in terms of generalization to new data. Feature importance analysis identifies satellite thermal variables, meteorological conditions, and proximity to infrastructure as having the greatest influence on predictions. These findings confirm that the integration of satellite-based and environmental multivariable data provides a more comprehensive foundation for supporting forest fire early warning systems and data-driven mitigation strategies.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Kebakaran hutan, Data Satelit, Regresi Hibrida, Integrasi Data, Genetic Algorithm, Forest Fires, Satellite Data, Hybrid Regression, Data Integration, Genetic Algorithm
Subjects:	H Social Sciences > HA Statistics > HA30.3 Time-series analysis T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T57.84 Heuristic algorithms.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 59101-(S2) Master Thesis
Depositing User:	Benedictus Augusta Vianney
Date Deposited:	27 Jan 2026 07:28
Last Modified:	27 Jan 2026 07:28
URI:	http://repository.its.ac.id/id/eprint/130517

Actions (login required)

View Item