Deep Learning based models for Forest Fires Prediction in Algeria

Abstract

Automatic prediction and detection of forest fires is an important field of research that contributes to disaster mitigation. Deep learning (DL) approach has emerged as a successful approach in forest fires detection and prediction, having a high accuracy rate with its distinctive learning mechanism. In this research, we have investigated the deep learning based prediction models for forest fires prediction using two models, namely the deep neural network (DNN) and the long short-term memory (LSTM), a variant of the recurrent neural network (RNN). The performance of the forest fire prediction DL-based models have been analyzed by computing the accuracy, recall and precision metrics. The new extended Algerian forest fires dataset is used in this study. The DNN and LSTM networks classify instances into two classes: fire and no fire. The results of the DNN and LSTM classifiers are also compared with the classical machine learning (ML) algorithms, namely the simple decision tree (DT), the support vector machine (SVM), the K-nearest neighbor (KNN), the Boosted tree and the Bagging tree. The proposed deep learning approach produced the best accuracy compared to the classical ML algorithms considering the two models, namely the DNN and the LSTM. The obtained classification accuracies are about 98.43% and 98% for the DNN and the LSTM models, respectively. The results of this study indicate that the DL-based classifiers improve the fire-occurrence prediction accuracy.