Causal Discriminative Modeling for Robust Cloud Service Fault Detection

Abstract

In this paper, a detection method based on a causal discrimination mechanism is proposed to improve the fault identification ability of the system in the complex dynamic environment to solve the modeling problem of service fault detection tasks in the cloud computing environment. This method explicitly models the causal dependency between multi-dimensional input features by constructing a structured representation module and causal path extraction mechanism and then generates intermediate representation with structural semantics for fault state discrimination. The overall architecture consists of input encoding, causal encoder, structure selector, and unified discriminator, which can effectively capture the key interaction paths between service components and enhance the perception ability of the fault propagation chain. In the process of structural modeling, the method combines causal graph information and structure selection function to extract stable and interpretable features for final prediction, which significantly enhances the robustness and transferability of the model under input perturbation and distribution change. In order to fully verify the effectiveness of the proposed method, this paper designs multiple experimental tasks, covering the sensitivity evaluation of multiple key variables such as the number of candidate paths, encoding dimension, noise perturbation, etc., and compares them with representative methods in recent years. Experimental results show that the proposed method performs well in multiple indicators such as accuracy, structural consistency, and stability, verifying the applicability and effectiveness of causal modeling mechanisms in cloud service fault detection.