Wildfires Classification Using Feature Selection with K-NN, Naïve Bayes, and ID3 Algorithms

Abstract

Wildfires are a problem with a high intensity of occurrence and recurrence in Indonesia. If this problem is not properly addressed, it will threaten air circulation in the world. The source of fire can be natural or man-made. As a preventive measure for the widespread spread of fire, it is necessary to investigate the type of fire early on so that it can be determined the type of fire with the highest priority to be extinguished immediately. The process of identifying fire types can be done by classification. This research aims to classify the type of fire with three algorithms, namely K-Nearest Neighbour (K-NN), Naïve Bayes and Iterative Dichotomise 3 (ID3). The forest fire dataset was obtained from the Global Forest Watch (GFW) platform. Before entering the classification stage, the dataset went through a feature selection process, where attributes meeting the threshold were selected for the classification process. The performance of ID3 algorithm is superior compared to other algorithms with an accuracy of 65.83, precision 67.4, recall 67.02 and F1 67.21 per cent. Finally, the feature selection process contributes positively to the classification process, increasing the model performance by 2-5 per cent.