A Distance-Based One-Class Classification with Bagging Ensemble for Fault Detection

A Distance-Based One-Class Classification with Bagging Ensemble for Fault Detection

This study introduces a novel distance-based Bagging ensemble one-class classification (DB-EOC) method for fault detection in avionics equipment. This approach overcomes the limitations of traditional one-class classification methods in terms of classification performance and generalization ability by defining a boundary between normal and abnormal instances using only normal sample data for training. By integrating multiple learners in parallel, the DB-EOC algorithm significantly enhances the model’s classification performance. Additionally, the paper explores the effectiveness of different distance metrics, including Euclidean distance, Manhattan distance, and Chebyshev distance, for fault detection. Experiments conducted on datasets from the University of California Irvine (UCI) and avionics equipment data demonstrate the efficacy of the DB-EOC method, particularly highlighting the superior performance of the Manhattan distance-based approach in fault detection accuracy. This research contributes to improving the reliability and safety of avionics systems by providing an effective fault detection strategy in scenarios where fault samples are scarce.