A novel classification method based on an online extended belief rule base with a human-in-the-loop strategy

Abstract

Classification methods, such as fault diagnosis and intrusion detection, are widely used in modeling complex systems. The accuracy and credibility of these methods directly affect the reliability of the modeling results, which in turn determines the effectiveness of engineering decisions. Additionally, the model's ability to be dynamically updated should be considered, given the intricate and ever-changing nature of engineering environments. For online models, adding new training samples without considering their suitability can lead to problems such as poor model performance and increased rule base complexity. Furthermore, amid constantly arriving new samples in a dynamic environment, modeling based only on initial expert knowledge can result in new samples not being fully used. Therefore, a novel classification method based on an online extended belief rule base with a human-in-the-loop strategy (OEBRB-H) is proposed in this paper. First, a fuzzy c-means algorithm based on expert knowledge (FBE) is designed to evaluate model parameters online. Second, a human-in-the-loop strategy for dividing the new sample set and a domain-value-based rule updating method are proposed for model optimization. Finally, two case studies, namely, aeroengine inter-shaft bearing fault diagnosis and industrial control intrusion detection, are performed. The results indicate that the model proposed in this paper can maintain both credibility and high accuracy in dynamic environments.