Prediksi Tingkat Bahaya Kemunculan Titik Panas Penyebab Kebakaran Hutan dan Lahan Menggunakan Metode Neural Network dan Random Forest

Sari, Devi Novita (2024) Prediksi Tingkat Bahaya Kemunculan Titik Panas Penyebab Kebakaran Hutan dan Lahan Menggunakan Metode Neural Network dan Random Forest. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003201066-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (7MB) | Request a copy

Abstract

Kebakaran hutan dan lahan merupakan masalah tahunan yang sering terjadi di wilayah sekitar garis khatulistiwa, termasuk Indonesia. Pada tahun 2023, tercatat 2.052 kejadian kebakaran hutan dan lahan di Indonesia yang mengakibatkan kerugian finansial, ancaman keselamatan, maupun gangguan ekosistem. Besarnya dampak dari kejadian tersebut mendorong peneliti untuk melakukan prediksi titik panas dengan memanfaatkan data penginderaan jauh (remote sensing). Penelitian ini berfokus pada identifikasi kemunculan titik panas (hotspot) di Kepulauan Bangka Belitung selama Juli hingga Desember 2023. Berbeda dengan penelitian lain yang umumnya hanya mempertimbangkan faktor iklim, penelitian ini juga mempertimbangkan pengaruh dari biofisik tanah terhadap kemunculan hotspot. Faktor iklim yang diduga mempengaruhi kemunculan hotspot yaitu temperatur maksimum, curah hujan, total penguapan, dan kelembapan spesifik, sementara faktor biofisik tanah meliputi indeks area hijau, tutupan lahan, dan jenis tanah. Variabel iklim dan biofisik tanah kemudian digunakan sebagai input untuk memprediksi tingkat hotspot menggunakan metode Random Forest dan Neural Network. Pemodelan klasifikasi dilakukan melalui tahap training dan testing menggunakan stratified k-fold cross validation dengan 5 fold, serta optimasi parameter menggunakan grid search. Hasil klasifikasi tingkat hotspot menunjukkan bahwa model Random Forest dengan pembobotan ’balanced’ memberikan prediksi yang lebih baik untuk kasus data imbalance dibandingkan dengan Neural Network, serta unggul untuk nilai AUC, balanced accuracy, dan specitifity. Prediksi data testing dengan Random Forest menghasilkan rata-rata AUC 0,641, G-Mean 0,405, balanced accuracy 0,596, sensitivity 0,25, serta specitivity 0,942. Metode Random Forest lebih baik dalam memprediksi hotspot tingkat sedang, namun kurang sensitif dalam memprediksi hotspot tingkat tinggi. Sementara itu, metode Neural Network lebih sensitif dalam memprediksi hotspot tingkat tinggi, namun lemah dalam memprediksi hotspot tingkat sedang.
==============================================================================================================================
Forest and land fires are annual issues frequently occurring in equatorial regions, including Indonesia. In 2023, there were 2,052 recorded fire incidents in Indonesia, leading to financial losses, safety threats, and ecosystem disruptions. The significant impact of these incidents has prompted researchers to predict hotspots using remote sensing data. This study focuses on identifying hotspot occurrences in the Bangka Belitung Islands from July to December 2023. Unlike other studies that generally only consider climatic factors, this research also considers the influence of soil biophysics on hotspot occurrence. Climatic factors suspected to influence hotspot occurrence is maximum temperature, rainfall, total evaporation, and specific humidity, while biophysical soil factors include green area index, land cover, and soil type. These climate and soil biophysical variables are used as inputs to predict hotspot levels using Random Forest and Neural Network methods. Classification modeling is performed through training and testing stages using stratified k-fold cross-validation with 5 folds, and parameter optimization using grid search. The hotspot classification results indicate that the Random Forest model with 'balanced' weighting provided better predictions for imbalanced data compared to the Neural Network, excelling in AUC, balanced accuracy, and specificity. Testing data predictions with Random Forest yield an average AUC of 0.641, G-Mean of 0.405, balanced accuracy of 0.596, sensitivity of 0.25, and specificity of 0.942. The Random Forest method was more effective in predicting moderate hotspots but less sensitive in predicting high level hotspots. Meanwhile, the Neural Network method was more sensitive in predicting high level hotspots but less sensitive in predicting moderate level hotspots.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Data Imbalance, Hotspot, Kebakaran Hutan, Neural Network, Random Forest, Forest Fire, Imbalance Data
Subjects:	G Geography. Anthropology. Recreation > GA Mathematical geography. Cartography > GA102.4.R44 Cartography--Remote sensing Q Science Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.6 Computer programming. Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) Q Science > QA Mathematics > QA76.F56 Data structures (Computer science) Q Science > QA Mathematics > QA9.58 Algorithms
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Devi Novita Sari
Date Deposited:	27 Aug 2024 05:37
Last Modified:	27 Aug 2024 05:37
URI:	http://repository.its.ac.id/id/eprint/115246

Actions (login required)

View Item