An Ensemble Fuzziness-Based Online Sequential Learning Approach and Its Application

Wei-Peng Cao, Sheng-Dong Li, Cheng-Chao Huang, Yu-Hao Wu, Qiang Wang, Da-Chuan Li, Ye Liu

Abstract: Traditional deep learning algorithms are difficult to deploy on most IoT terminal devices due to their limited computing power. To solve this problem, this paper proposes a novel ensemble fuzziness-based online sequential learning approach to support the local update of terminal intelligent models and improve their prediction performance. Our method consists of two modules: server module and terminal module. The latter uploads the data collected in real-time to the server module, then the server module selects the most valuable samples and sends them back to the terminal module for the local update. Specifically, the server module uses the ensemble learning mechanism to filter data through multiple fuzzy classifiers, while the terminal module uses the online neural networks with random weights to update the local model. Extensive experimental results on ten benchmark data sets show that the proposed method outperforms other similar algorithms in prediction. Moreover, we apply the proposed method to solve the network intrusion detection problem, and the corresponding experimental results show that our method has better generalization ability than other existing solutions.