Computing and Informatics

Stock prediction is prominent in the field of Artificial Intelligence. Stock prediction problems are handled either as a regression or classification task. Studies in the literature have also shown success for hybrid learning to stock prediction. But little attention is paid to finding out the effect of spatial feature extraction/distortion over the temporal effect of the deep neural network and vice versa for the problem under study. The paper, therefore, proposes a hybrid long short-term memory (LSTM) network over a convolutional neural network (CNN) called LSTM-CNN as against the popular CNN-LSTM model. The daily price movement of the S & P 500 index data is utilized. A sliding window technique is considered to obtain a balanced data of 20-days window data from the S & P 500. The proposed stock prediction model is investigated further for an optimal set of hyperparameters using the Bayesian optimization (Bo) technique. In addition, the proposed model is compared with optimized CNN, LSTM, and CNN-LTSM models. The optimized LSTM-CNN model is found to outperform the other models with accuracy, precision, and recall values of 0.9741, 0.9684, and 0.9800, respectively. The proposed model is established to provide a better stock trend prediction.