A dynamic pattern recognition approach based on neural network for stock time-series

Pattern theorem in financial time-series is one of the most important technical analysis methods in financial prediction. Recent researches have achieved big progresses in identifying and recognizing time-series patterns. And most of the recent works on time-series deal with this task by using static approaches and mainly focus on the recognition accuracy, but considering that recognition of patterns in financial time-series, especially for stock time-series, are always time-consuming rather than pattern recognition in other fields, a dynamic recognition approach is more preferable so that investment on stock pattern become executable. In this paper we propose a dynamic approach for extracting and recognizing the patterns in stock-series. In our approach a slide window with flexible length is defined for extracting feature vertexes in stock-series, and in addition, a dynamic perceptual important point (PIP) locating method is proposed based on the PIP locating method for avoiding the computation expense problem and an artificial neural network (ANN) approach is involved for pattern recognition and window length identification.