An FDA-based Stock Exchange Price Curve Feature Recognition and Analysis Method

摘要

The classification and trend prediction of stock exchange price via trading history becomes the crucial part of intelligent stock analysis software nowadays. To figure out the variation pattern of stock price better, the curve-based method is proved to be efficient when applied to large discrete dataset. This paper proposed a FDA-based stock price curve recognition method in order to provide support for stock price prediction. On the basis of fitting function, extract segmented variation trend, segmented variation rate and segmented Root Mean Square as features which reflect the information of curve shape. And give weight to the three features to form the feature vector of the curve. Then conduct K-means clustering on these feature vectors. The result is the same to the subjective classification, so that in this way obtaining the class label of each curve. Finally classify the unknown curve with neural network. On the test set, the correct recognition rate reaches 80%.