Incorporating an attention layer in deep learning models to enhance short-term stock price predictions

Abstract In recent years, deep learning models have gained significant attention in the field of stock market prediction. However, accurately forecasting short-term stock prices remains a challenging task due to the inherent volatility and complexity of financial markets. RNN Algorithms like LSTM (Long Short-Term Memory) and GRU (Gated Recurrent Unit) provide decent predictions on stock market time series data. However, these algorithms face bottleneck situations that arise due to the complex nature of the input sequences. This research paper aims to investigate the effectiveness of incorporating attention layers in deep learning models to improve the accuracy of short-term stock price predictions. Subsequently, the layer assigns weights to the attributes based on their scores, highlighting the attributes that are more relevant within the sequence. With the integration of attention layer into the encoder-decoder model, we are able to extract and emphasize pertinent information from the complex time series sequences, thereby enhancing the accuracy of hybrid to 95% and attention reached to 99% accuracy.