Stock Price Fluctuation Prediction Based on Text Analysis of Stock Market News

Abstract

This thesis is presented with a prediction of stock market price change using news and announcements. Stock Market News provides invaluable information to brokers and investors to take their crucial decisions on investing in stock market. Most of modern stock market data dissemination software systems and tools provides various types if indicators to facilitate these decisions. It includes real-time indicators and historical graphs of stock prices. Adding more indicators to this list is an added advantage to any of those systems due to the impact on stock market activities. There are several researches done on same subject by several people. However, most of those investigations were limited to a specific region in the world (US stock markets), specific resources (Yahoo finance, Google trends etc) and specific prediction parameters (price trend positive or negative). The aim of this project is to build a data model and prediction system which facilitate another dimension of indicators. It would provide information on the stock price trend for few days ahead, based on current stock market news information. This research includes enhanced prediction methodology to existing methods by providing additional information such as the number of days that the trend will be retained. Since this is a text analysis-based project, few classification algorithms have been used. Weka tool has been used to feature extraction process and classification. Results were obtained as a comparison between each algorithm used. Best training data set was identified using K-Fold Cross Validation techniques applied to all algorithms. The best results were obtained using Random Forest algorithm with a significant accuracy level.