Hotspot Prediction Using 1D Convolutional Neural Network

Abstract Every year, there are always forest and land fires in Indonesia. Riau Province is one area that often experiences forest and land fires. One of the factors causing forest fires is the emergence of hotspots. The hotspot is an indicator of forest fires that indicates a location has a relatively higher temperature than its surroundings is ≥42°. This research studies about hotspot prediction using a 1d convolutional neural network to give some information about the number of hotspots for early treatment of forest fires. Experiments carried out by changing the amount of output that is daily, monthly, and 12 months. It also experimented on the learning rate and number of nodes used on neural networks. The results of this research are using 20 nodes and a learning rate of 0.2 on daily predictions with a MAAPE of 0.9386, 40 nodes and a learning rate of 0.5 on a monthly prediction with a MAAPE of 0.4139, 40 nodes and a learning rate of 0.5 on a prediction of 12 months with a MAAPE of 0.4397.