Stock Forecasting for Time Series Data using Convolutional Neural Network

Deep learning approach plays vital role in prediction of financial time series data. In this paper, a linear autoregressive model (AR) is trained from the historical data and stochastic term to predict future outcome. The Korean stock market time series data consists of 1239 days and trading observations have 5 min of informational data for analysis of stock market prediction. The applied model normalizes the data for prediction interval of future returns, and generates efficient forecast. Autoregressive model generates less error in comparison to logistic regression. Further, the rises and falls in stock prices are classified using a functional Link artificial neural network (FLANN) and convolutional neural network (CNN). The comparative analysis and observations states that CNN performs better providing an accuracy of 93.24 % during training and 97.56 % during testing phase whereas FLANN shows accuracy of 90.26% during training and 95.90% during testing.