Stock Price Prediction using Depthwise Pointwise CNN with Sequential LSTM

Recent developments in computing technology have resulted in the continuous accumulation of enormous volumes of stock data and information. Due to the market's volatile, unpredictable, and non-stationary nature, analyzing stock market movements and price behavior is extremely difficult. In this study, stock prediction model is proposed using a recurrent neural network with depthwise separable convolution for smoothing the prediction. Depthwise separable convolution is used in the proposed model to improve feature extraction. Extracted features are provided in sequential LSTM to forecast future price of the stock. The proposed model., Depth-wise Separable CNN with Sequential LSTM (DWCNN-SLSTM) is evaluated on S&P 500, HSI, CSI300, and Nikkei 225 datasets. The proposed model outperforms the existing and achieved MAPEof 0.4734, 0.5051, 0.4865, and 0.4776 on S&P500, HSI, CSI300, and Nikkei 225 respectively.