PREDICTIVE MODELING OF FOREST FIRES USING MACHINE LEARNING

Abstract

Forest fires are among the worst natural disasters, causing extensive damage and loss to various forms of life, including humans and infrastructure. In recent years, these catastrophes have been occurring more frequently and without warning, highlighting the urgent need for more intelligent systems to predict and manage the impacts of climate change. This thesis presents a method based on machine learning that analyzes meteoro logical and environmental data to determine the likelihood of forest fires. The study utilizes techniques derived from Artificial Neural Networks (ANNs), incorporating ge ographic, weather, and temporal data to create a reliable prediction tool. The dataset used contains a total of 518 instances with variable features such as temperature, wind speed, humidity, rainfall, and Fire Weather Index component data. The training and evaluation of the model were carried out using these features. Proper data preprocessing, including normalization and model optimization techniques, significantly improved the classifier’s performance. The proposed model achieved a prediction accuracy of 96%, surpassing several standard machine learning algorithms. This study compares the performance of the proposed model with algorithms such as Support Vector Machines (SVM), K-Nearest Neighbors (KNN), and Decision Trees. The results indicate a strong potential for AI-powered systems to play a meaning ful role in understanding environmental hazards, promoting timely actions, informed policies, and efficient use of resources. Overall, the findings of this research contribute to disaster management by offering a flexible and accurate model that aids decision-makers in effectively controlling forest fires and shaping future strategies in the field.