Extracting the best features from multi-company stock data to improve stock price prediction
Bonde, Ganesh Suresh
MetadataShow full item record
To predict stock price is always a challenging task. The stock prices are dependent on many factors. Much research has been done in this field but it has been difficult to find the right features, which will help predict the prices with high accuracy. When you try to analyze stock price using historical stock data, it has many attributes (feature set) containing useful as well as redundant features. Thus there is a need to remove the unwanted stock information. In this research we first try to find the best features from the available company data as well as data about similar companies and stock indexes. The best-extracted features are then used to predict stock prices using different machine learning algorithms. Based on the results obtained in the previous experiments, we then implemented two new techniques for predicting stock prices. We used genetic algorithms and evolution strategies. The results obtained using these algorithms were promising. In each case the accuracy obtained was more than 70%. In this research, data of eight companies was used, each having six attributes. Also NASDAQ and S & P 500 data was used for predicting the stock prices.