Towards A Deep FLANN For Prediction Of Stock Market Returns

Stock prediction of an individual company is an important and difficult problem for wealth management. Many techniques have been developed for analysis of stock market, however, it remains a challenging problem. One approach is to use deep learning for stock price prediction and some success has been reported in literature with Convolutional Neural Networks (CNN). This paper proposes deep functional link neural network for stock predictions as functional link artificial neural networks (FLANN) are single-layer neural networks which can solve complex problems by generating nonlinear decision boundaries. It is conjectured that deep FLANN would benefit from the ability of deep learning to handle complex problems generated from multiple hidden layers and interconnection of neurons as well as from the inherent capability of single layer FLANN to handle difficult problems. Initial investigations show that average performance of Deep FLANN is better than the single-layer FLANN as well as CNN for prediction of stock prices of a set of companies listed on Korean Stock market.