An Abnormal Behavior Detection Model Based on Expert Routing and Machine Learning

This paper introduces a model FFA-Expert based on expert routing network, which aims to improve the accuracy of abnormal behavior detection by integrating multiple machine learning techniques. FFA-Expert combines five machine learning models with different weight configurations to create a collection of expert models. During data processing, the routing component activates K experts instead of utilizing all models, thus reducing the computational overhead. In addition, the traditional sequence data is converted into 3D format and trained with the sequence data to maximize feature extraction. Aiming at the time uncertainty of dynamic data set, this paper adopts dynamic update strategy and adjusts the parameter update according to the optimization direction of model weight. The experiment proves that FFA-Expert outperformed machine learning models including the current mainstream in terms of AUC and recall metrics. The model shows particular effectiveness in identifying abnormal behavior, but also highlights its robustness in pattern recognition. The research results not only provide a new way to improve the detection efficiency in various fields, but also provide a meaningful direction for future research.