Nonnegative matrix factorization-based method for detecting anomalies in stock market

Abstract

The invention relates to a nonnegative matrix factorization-based method for detecting anomalies in the stock market. The method includes the following steps that: a stock index matrix is built; the stock index data matrix is factorized through using NMF (nonnegative matrix factorization), so that a base matrix U representing stock index feature bases and a coefficient matrix V representing low-dimension weight coefficients are obtained; nonnegative matrix factorization is performed on a stock time sequence Xn*m, so that a base matrix Un*r and a coefficient matrix Vr*m can be obtained; wavelet transformation is performed on a weight coefficient vector Vi, so that multi-level waveforms of different granularities are obtained; the fluctuation amplitude of the waveforms is detected, anomaly conditions are judged according to the amplitude of the waveforms; and after an anomaly position of the sequence is determined, empirical analysis is carried out: the detected anomaly fluctuation position relative to the sequence is found out in the sequence of which the weight coefficient vector Vi has been subjected to wavelet transformation, the time point of an abnormal event is marked in a corresponding position in original matrix data, the transformation conditions of stock market indexes at the time point are inspected, and therefore, the accuracy of detection is judged.

Description

technical field [0001] The invention belongs to an intelligent detection algorithm of a stock market, in particular to an abnormality detection method in a stock market based on non-negative matrix decomposition. Background technique [0002] Anomaly detection aims to detect data that does not meet the expected behavior, so it is suitable for many fields such as fault diagnosis, disease detection, intrusion and fraud detection, and financial market fluctuation detection. In the research on theories and models related to stock market volatility, it mainly focuses on the analysis of abnormal fluctuations. Abnormal fluctuations in stock market data with time series usually lead to model parameter estimation deviations, low volatility prediction accuracy, and conclusions. Some invalid conclusions etc. Therefore, it is of great significance to detect outliers in the time series data of the stock market. [0003] Generally, according to the theory and method of machine learning ...

AI Technical Summary

Problems solved by technology

[0004] Bilen C and Huzurbazar, S.Grane proposed a wavelet-based outlier detection method, but the average error rate of outliers detected by this method is high

Images