Converging Deep Learning Neural Network Architecture for Predicting NSE-50

The proposed method of stock rate prediction is based on the general idea, “current news has much influence on stock price movements”. News of the selected stocks has been taken from two important sources - moneycontrol.com and daily newspaper, Economics Times. A fuzzy mathematical value between 1 to -1 has been calculated and used as input to the proposed system. Stocks & Other input values are normalized stock price and volume data for long-term, medium-term and short-term prospective. Data of Top 50 companies from National Stock Exchange (NSE) has been taken for the duration of 5 years. Converging Deep Learning Neural Network (CDL-NN) Architecture has been developed and trained with Error Back Propagation algorithm has been used. The results are compared with a two hidden layer neural network results on the same data. The proposed model produced 2 to 3 percent better results for NSE top 50 stocks.