Anomaly detection based on Non-negative matrix factorization in stock market

The anomalous fluctuations of the stock market should affect the normal operation of the whole financial market, trigger the release of all kinds of unstable factors in financial operation. Therefore, we can quickly find potential risks in the market by analyzing the data of the stock market and detecting anomaly fluctuations. Due to the stock market to produce huge amounts of data, the data processing power demand is high. So detecting anomalous point fast, efficiently and accurately is important the research direction. In the article, we consider the time sequence characteristics of stock data, use the method of non-negative matrix factorization(NMF)to acquire weight coefficient set that denote the characteristic of stock date perfectly .Then these weight coefficient set is decomposed by wavelet transform, so, we can calculate the abnormal fluctuations from all of decomposition hierarchy. Finally, anomaly fluctuations can be quickly and accurately positioned out by means of weighted fusion. In the process of anomalous fluctuations detection, we introduce the weight fusion means, implement two weight, and find out abnormal points finally. In this way, more accurate detection result can be obtained. The empirical analysis found that our method is coincident with real condition, it can ensure higher detection accuracy besides persistent fluctuation in certain time period. Keyword: Anomaly detection; Non-negative matrix factorization; Data mining; Feature extraction