FWRF: Mitigating Redundant Features in Fault Diagnosis via Feature Weighted Random Forest

The performance of data-driven fault diagnosis methods is often impeded by irrelevant and redundant information in high-dimensional data. To address this issue, this article proposes a fault diagnosis method based on feature-weighted-inspired random forest (FWRF). In this method, a novel Boruta recursive feature elimination with cross-validation (BRFECV) algorithm is proposed to obtain an optimal feature set with associated weights. Moreover, a novel FWRF architecture is developed, replacing simple random sampling in random forest (RF) with weighted random sampling for feature subset formation. The assignment of weights to each feature is determined based on a specifically designed feature evaluation approach. Consequently, the proposed FWRF method can effectively use feature information while reducing noise interference. Finally, the proposed FWRF-based method is implemented for online fault diagnosis, with the effectiveness of the proposed method validated using satellite attitude control system data from a semi-physical simulation platform. Experimental results demonstrate that the proposed approach exhibits superior practicality and performance in terms of accuracy, running time, and F1-score compared with other methods for fault diagnosis in satellite attitude control systems.