Deep Data Stream Analysis Model and Algorithm With Memory Mechanism
Kun Gao, Yiwei Zhu · IEEE Access · 2016
Integrated analysis is an important method for data analysis. Aimed at improving the deficiencies of traditional integrated data stream analysis, a human-like remembering and forgetting mechanism is introduced into data stream analysis, and a deep data stream analysis model based on remembering (DSAR) is proposed. Through this remembering and forgetting mechanism, the model regards basic classifiers as system-obtained knowledge and not only stores useful basic classifiers in a “remembering library” to improve prediction stability but also selects good basic classifiers to participate in integrated prediction, thus improving its ability to accommodate conceptual variations. Based on the DSAR model, an integrated deep data stream analysis (DDSA) algorithm is proposed. The algorithm uses the forgetting curve and a selective ensemble classifier to simulate human thinking. Compared with four typical data stream analysis algorithms, the DDSA algorithm has a high classification accuracy and a strong capacity for accommodating concept drift features (CDFs) within data stream analysis. The DDSA is particularly adaptable to complex CDFs in practical applications. Experiments show that the proposed algorithm can not only adapt to new concept changes quickly but also effectively resist the impact of random fluctuations on system performance.