Imbalanced Data Fault Diagnosis Based on Hybrid Feature Dimensionality Reduction and Varied Density-Based Safe-Level Synthetic Minority Oversampling Technique

This chapter proposes an imbalanced data processing method that combines hybrid feature dimensionality reduction technology with varied density-based safe-level synthetic minority oversampling technique (VDB-SLSMOTE). Firstly, a hybrid feature dimensionality reduction method is utilized to reduce the dimensionality of high-dimensional unbalanced data, based on compensated distance evaluation technique (CDET) to select sensitive features from high-dimensional features, and kernel principal component analysis (KPCA) to transform the data features to low-dimensional features. Secondly, a VDB-SLSMOTE algorithm is proposed in this chapter to solve the intra-class imbalance problem. The varied density-based spatial clustering of applications with noise (VDBSCAN) method is used to cluster the minority samples with the intra-class imbalance and the SLSMOTE algorithm is used to synthesize new samples in each cluster. Finally, the effectiveness of the proposed algorithm is verified by applying this method to the diagnosis of imbalanced planetary gearbox gear fault data.