Machine Learning for Big Data Analytics in Development of Wildfire Prediction Models

This study aims to develop a model that predicts domestic forest fire occurrences during fire outbreaks using machine learning techniques. For the modeling methods, logistic regression analysis and ensemble techniques, such as gradient boost and random forest, were used while the oversampling technique was utilized to address the imbalance problem of the forest fire data. The model developed in this study predicted 239 out of 333 forest fire occurrences during the nationwide forest fire period in 2020 with a prediction accuracy of approximately 71.8%. Forest fires that occur during such periods are highly influenced by different factors affecting the climate, such as temperature, humidity, and precipitation. In Gangwon-do, in addition to these factors, a high correlation between farmland density and stem volume per hectare has also been associated with increased forest fire occurrences. The significance of this study lies in the fact that it presents a customized wildfire occurrence prediction model that can be used in the administrative parts, which serve as the basic centers for wildfire prevention, of provinces and cities across the country.