Novel deep learning model with fusion of multiple pipelines for stock market prediction

Deep learning has become a powerful tool in modeling complex relationships in data. Convolutional neural networks constitute the backbone of modern machine intelligence applications, while long short-term memory layers (LSTM) have been widely applied towards problems involving sequential data, such as text classification and temporal data. By combining the power of multiple pipelines of CNN in extracting features from data and LSTM in analyzing sequential data, we have produced a novel model with improved performance in stock market prediction by 20% upon single pipeline model and by five times upon support vector regressor model. We also present multiple variations of our model to show how we have increased accuracy while minimizing the effects of overfitting. Specifically, we show how changes in the parameters of our model affect its scores for training and testing, and compare the performance of a multiple pipelines model using three different kernel sizes versus a single pipeline model.