A Study on Stock Data Mining by Map Recognition

Stock data analysis for price forecasting and trend prediction has been a challenging problem that attracts researchers from different fields. Some use statistical methods, while others use neural network based approaches. This paper reports on a preliminary study on stock market data analysis using a hyperspace data mining approach that is built upon a projective geometrical method. Discussions include data separation, feature selection, data pattern identification, and model building. Application of this method to stock performance classification and market speculation prediction is described. Preliminary results with real-world financial data seem to provide useful insights on how to discriminate the performance of different companies and to identify the market speculation manipulated by large investors.