In this project, I build and compare four types of ANN models: fully connected ANN, RNN, LSTM, LSTM with Attention. There are two datasets which contain time series. The goal is to build deep neural networks which can learn the temporal patterns in data and predict a value of future observation. For those models, I compare the accuracy of predictions and the speed of the training process. Please refer to Report.pdf for detailed description and references.

To build the neural networks I use python keras library. To implement Attention Mechanism I used the source code of Christos Baziotis.