Integrated financial data stream mining based on memory curve

Integrated financial data stream mining is an important method to study financial data stream with the sample drift. In view of the defects of the traditional integrated financial data stream mining, human memory curve is introduced into the financial data stream mining, and the memorizing based financial data stream mining (MFSM) is proposed. The model regards the base classifier as the knowledge gained by the system, through the Memory mechanism, it not only makes the base classifier useful in history be kept in the Memory attributing to the strong memory capacity, and improve the stability of the prediction, but also select the base classifier with good classification effect from the Memory bank to participate in the integrated prediction, so as to improve the adaptability to the sample variation. Based on the MFSM model, a memorizing based integrated financial data stream mining (MISM) is put forward, which makes use of the Ebbinghaus memory curve to design the memory curve of the system, and utilizes the selective integration to simulate human's Memory mechanism. After compared with 4 typical financial data stream mining algorithms, the results show that: The MISM algorithm has high classification accuracy and strong overall adaptability to the sample drift, especially for the sample drift that ocurrs repeated and the complex sample drift existing in the practical application. It not only can adapt to the new sample variations quickly, but also can effectively resist the impact of the random sample fluctuations on the system performance.