One of the most concentrated areas of time series assessment is anticipating the expense of stocks. With propels in AI and profound learning innovation, the examination of monetary information for stock cost anticipating has altogether sped up. RNN algorithms like LSTM (Long Short-Term Memory) and GRU (Gated Recurrent Unit) provide precise projections on stock market time series data. These calculations, then again, face bottleneck issues because of the intricacy of the information arrangements. The issue is then resolved by incorporating an attention layer into the deep learning model. It is used to solve the bottleneck problem that arises in encoder–decoder systems during the text analysis of sentences of varying lengths and lengths. In order to get attributes that are more relevant to the outcome, we have added an attention layer to our Bi-LSTM and GRU models in this paper. The attention layer scores each attribute based on how closely the input sequence matches the current states of the encoders and decoders using the output vectors from the encoders and the hidden states of the decoders. The layer then, assign weights to the attributes based on the scores to highlight which attribute is more relevant in the sequence. Thus, by adding the attention layer to the encoder–decoder model, we can highlight relevant information from the complex time series sequences and hence provide better stock price predictions. 

Katja Koch

Papers

Improve Short-Term Stock Price Forecasts Through Deep Learning Algorithms