BiCuDNNLSTM-1dCNN — A hybrid deep learning-based predictive model for stock price prediction

Within last decade, the investing habits of people is rapidly increasing towards stock market. The nonlinearity and high volatility of stock prices have made it challenging to predict stock prices. Since stock price data contains incomplete, complex and fuzzy information, it is very difficult to capture any nonlinear characteristics of stock price data, which usually may be unknown to the investors. There is a dire need of an accurate stock price prediction model that could offer insights to the investors on stock prices, which ultimately could deliver positive investment returns. This research is focused on proposing a hybrid deep learning (DL) based predictive model, that combines a Bidirectional Cuda Deep Neural Network Long Short-Term Memory (BiCuDNNLSTM) and a one-dimensional Convolutional Neural Network (CNN), for timely and efficient prediction of stock prices. Our proposed model (BiCuDNNLSTM-1dCNN) is compared with other hybrid DL-based models and state of the art models for verification using five stock price datasets. The predicted results show that the proposed hybrid model is efficient for accurate prediction of stock price and reliable for supporting investors to make their informed investment decisions. • BiCuDNNLSTM-1dCNN is a hybrid DL model based on Bidirectional CuDNNLSTM and CNN. • BiCuDNNLSTM-1dCNN is efficient and scalable in developed and emerging stock market. • BiCuDNNLSTM-1dCNN uses univariate time series data to predict stock price. • Results confirm BiCuDNNLSTM-1dCNN is effective for volatility of stock price data. • BiCuDNNLSTM-1dCNN predicts better than LSTM, LSTM-CNN, CuDNNLSTM and LSTM-DNN.