Earthquake-prediction-using-LSTM

The reason for using this algorithm is that this can deal with the vanishing gradient problem in a better way than any other algorithm. Vanishing gradient problem may occur when we use large datasets and cause the gradients of the loss function (in our case we used mean square error) to decrease and it may approach zero which makes the network hard to train. When using any other ordinary neural net, the data in huge quantity can cause a decrease in the accuracy, neurons learn slowly in starting layers as compared to the last one. The type of model we have used is sequential and to tackle this heuristic problem. It performed well on time series data. So, if we want an algorithm that understands data set thoroughly, we’d use an LSTM which is a type of recurrent neural network. Simple DNN (deep neural network) and CNN (convolutional neural network) cannot be used in this scenario as it cannot deal with datasets with large timespan because they don’t have any memory. But RNNs, are a type of neural network, that can recall past training information. A regular recurrent network’s memory becomes more delicate as they fade into the past since the error signal from later time steps doesn't make it far enough back in time to influence the network at an earlier time. Memory matters when we have time-series data of a long period that’s why we'll use an LSTM which is an advanced form of recurrent neural networks. In a recurrent neural network, the hidden layer is a combination of the input data at the current time step and the hidden layer at a previous time step. The hidden layer is constantly changing as it gets more inputs and the only way to reach these hidden states is with the correct sequence of inputs this is how memory is incorporated in it and they're able to connect previous data with the present task.