Penerapan Support Vector Machine untuk Klasifikasi Tingkat Risiko Kebakaran Hutan

Abstract

Classifying forest fire risk levels is a critical step for disaster mitigation, yet it poses significant challenges due to data complexity and class imbalance. This study systematically applies and evaluates the performance of the Support Vector Machine (SVM) algorithm for the multi-class classification of fire risk (‘Low’,’Medium’,’High’) using the standard UCI Forest Fires dataset. The methodology involved a comprehensive preprocessing imbalance and hyperparameter optimization of C and gamma using GridSearchCV with cross-validation. Experimental results show that the final,optimized SVM model only achieved an accuracy of 50% and a macro-average F1-Score of 40% on the test set. This limited performance, particularly the model’s failure to reliably identify the ‘High’ risk class, indicates that the standard meteorological features within the dataset possess insufficient predictive power for the complex task of classifying fire severity, highlighting that model success is fundamentally dependent on feature richness over algorithmic optimization.