The Analysis and Forecasting of Stock Price with Deep Learning

Stock forecasting is one of the most popular topics nowadays. The dynamic, noisy and long-term dependence of stock market data makes its future prediction more difficult. This requires the use of additional data for successful prediction. In this study, the closing values of the stock data are predicted on a weekly basis by using the extended data set using various technical indicators and different independent variables. AAPL, NVDA, and GOOG stocks in the NASDAQ index were studied for the experiments. 20 different technical indicators obtained from daily stocks; different feature selection techniques were applied and then used as a feature vector for each day of the data. With the calculated technical indicators, a high dimensional feature space was created for data points that normally cover noise. We compare a multi layered Convolutional Neural Network (CNN) model, which we believe has achieved consistent results for prediction stock closing values, as well as a Long Short-Term Memory with Peephole (LSTMP) approach, which can cope well with long-term dependencies such as stock market data.