Predicting wildfires in Algerian forests using machine learning models

• DOI: 10.1016/j.heliyon.2023.e18064
• Publication Date: 2023-07-10T17:06:54Z
• Authors: Abdelhamid Zaidi
• Keywords: H1-99; 0106 biological sciences; Artificial intelligence; Science (General); Principal component analysis; Cross-validation; Social sciences (General); Q1-390; Machine learning; 0202 electrical engineering, electronic engineering, information engineering; Wildfire forecasting; Research Article

Algeria is one of the Maghreb countries most affected by wildfires. The economic, environmental, and societal consequences these fires can last several years after wildfire. Often, it possible to avoid such disasters if detection outbreak fire fast enough, reliable, early. lack datasets has limited methods used predict wildfires in mapping risk areas, which updated annually. This study result availability a recent dataset relating history forest cities Bejaia Sidi Bel-Abbes during year 2012. being small size, we principal component analysis reduce number variables 6, while retaining 96.65% total variance. Moreover, developed an artificial neural network (ANN) with two hidden layers cities. Next, trained compared performance our classifier those provided Logistic Regression, K Nearest Neighbors, Support Vector Machine, Random Forest classifiers, using 10-fold stratified cross-validation. experiment shows slight superiority ANN others, terms accuracy, precision, recall. Our achieves accuracy 0.967±0.026 F1-score 0.971±0.023. SHAP technique revealed importance features (RH, DC, ISI) predictions model.