A Deep Neural Network based model to forecast Geomagnetic Storm and observe the effects of various activation functions on the model

A solar flare and the energy exchange from the sun cause a geomagnetic storm, also known as a magnetic storm, on Earth's surface. Satellite and communication systems can be hampered during a magnetic storm. The DST index measures the magnetic activity that leads to magnetic storms. LSTM and GRU are two examples of cutting-edge machine learning algorithms that can be used to forecast the DST index. We proposed an artificial neural network-based model to forecast geomagnetic storms. The main layers of the base model were the LSTM layer, the convolutional layer, and the dense layer. We employed four activation functions to observe the effect of activation functions on the results of the model. The performance of the swish activation function was superior, while ELU was inferior. This work demonstrated the accuracy and loss of the model while different activation functions were used. The model error and validation error of our best evaluation were 0.1591 and 0.4658, respectively. We used RMSE to measure the performance of the model. The RMSE value of the best model was 0.4246.