High-Dimensional Feature Fault Diagnosis Method Based on HEFS-LGBM

Abstract

The challenge caused by redundant feature interference in high-dimensional fault feature data of analog circuits, will undermines the efficacy of conventional analog circuit fault diagnosis techniques, Thus, a novel approach termed Heterogeneous Ensemble Feature Selection (HEFS) is proposed in this paper. This approach is synergistically integrated with the Light Gradient Boosting Machine (LGBM) for pattern recognition, facilitating the prioritization and selection of significant high-dimensional features in analog circuit test data before classification. The methodology commences with the deployment of a heterogeneous ensemble learning strategy for the discernment of crucial high-dimensional features based on their significance. This is followed by the application of the LGBM technique for the pattern recognition classification of the earmarked features. Furthermore, the Tree-structured Parzen Estimator (TPE) optimization method, and five-fold cross-validation, are used for hyperparameter optimization to improve the model’s performance. Diagnostic evaluations are conducted on both University of California Irvine (UCI) datasets and analog circuits to underscore the superior diagnostic precision of the proposed HEFS-LGBM method compared with the existing techniques.