An Ensemble Stock Predictor and Recommender System

Abstract

Stock market analysis and prediction tools have been prevalent for several years now with various techniques and models to predict stock markets efficiently. This paper presents the design and implementation of a novel technique to predict stock market trends. The approach is an ensemble model which takes into account historical stock data, tweets and news affecting the stock prices of various companies and provides recommendations on which stocks to invest in for a particular duration. The model is built on real historical stock data set obtained from Indian National Stock Exchange (www.nseindia.com) over a 20-year period. The model uses Recurrent Neural Networks (RNN) and Long-Short Term Memory (LSTM) to learn and predict the future stock trend for a basket of 50 companies with RMSE of 1.43 (stock units with respect to Indian currency). Finally, the predicted stock price within a duration is converted into a graphical image, which is passed to a Convolutional Neural Network (CNN) classifier. The classifier is trained on finer features like peaks and troughs within the stock trend image, to provide recommendations on when to invest in a particular company stock. When tested with real stock prices over a week, it was found that the model was able to achieve extremely high accuracy in predicting the stock trends.