Classification of Forest Fire-Prone Areas Using the K-Nearest Neighbor Algorithm: A Case Study of Baluran National Park

Abstract

Forest and land fires are one of the most recurring and destructive natural disasters in Indonesia, particularly during the dry season when rainfall is significantly low. Among the most affected areas in East Java is Baluran National Park, a region highly vulnerable due to its dominant savanna ecosystem. In response to the urgent need for effective forest fire risk prediction, this study aims to classify fire-prone areas using weather-related data and machine learning techniques. The research focuses on the application of the K-Nearest Neighbor (K-NN) algorithm to predict fire risk levels based on meteorological parameters. The dataset used in this study was obtained from Visualcrossing.com and consists of 211 weather records with 32 explanatory variables, such as maximum and minimum temperature, wind speed, sea-level pressure, solar radiation, and solar energy, along with one target variable representing fire risk level (categorized as High, Medium, or Low). The research method involves several stages: data preprocessing (handling missing values and converting nominal data into numeric), transformation (splitting into training and testing sets using the Percentage Split technique), and classification using K-NN implemented on Google Colab. The K-NN algorithm was configured with K = 3, using Euclidean Distance as the distance metric. The classification process produced a high accuracy rate of 98%, indicating the robustness and effectiveness of K-NN in classifying forest fire risks based on weather data. This model was further validated by comparing predicted outputs against actual values in the testing dataset, showing high consistency. The results suggest that the K-NN algorithm is highly applicable for environmental classification problems and can support decision-making systems in early warning and disaster mitigation efforts. This study contributes to the growing field of data-driven disaster risk management and highlights the potential of machine learning in enhancing environmental monitoring systems.