New Efficient Wildfire Prediction using LSTM, CNN and Pretrained Models

The recent increase in wildfire outbreaks calls for efficient forecast and monitoring for minimizing ecological damage and controlling the adverse effects on human society. Machine Learning (ML) has been identified as an efficient tool for wildfire prediction. In this paper we propose new, efficient wildfire prediction methodologies by employing different deep learning techniques, like Long Short Term Memory (LSTM) network, Convolution Neural Network (CNN) and pretrained Residual Network (ResNet) and Densely Connected Convolution Network (DenseNet) models. After training and validating all our deep learning models over publicly available fire datasets, test results demonstrate that our designed LSTM, CNN, ResNet and DenseNet models achieve a good testing accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$76 \%, 91 \%$</tex>, 88 % and 96 % respectively. The models can successfully classify majority of the wildfire data, resulting in high true positive and true negative values.