[Paper Review] An Adaptive Oversampling Learning Method for Class-Imbalanced Fault Diagnostics and Prognostics

Wenfang Lin, Zhenyu Wu+1 more|arXiv (Cornell University)|Nov 19, 2018

Imbalanced Data Classification Techniques4 references4 citations

TL;DR

This paper proposes EWMOTE, an adaptive synthetic oversampling method that improves fault diagnostics and prognostics in class-imbalanced industrial systems by combining weighted minority sampling with EM-based imputation to generate high-quality minority-class samples. It achieves superior performance over baseline methods in both binary and multi-class fault classification tasks across varying imbalance ratios.

ABSTRACT

Data-driven fault diagnostics and prognostics suffers from class-imbalance problem in industrial systems and it raises challenges to common machine learning algorithms as it becomes difficult to learn the features of the minority class samples. Synthetic oversampling methods are commonly used to tackle these problems by generating the minority class samples to balance the distributions between majority and minority classes. However, many of oversampling methods are inappropriate that they cannot generate effective and useful minority class samples according to different distributions of data, which further complicate the process of learning samples. Thus, this paper proposes a novel adaptive oversampling technique: EM-based Weighted Minority Oversampling TEchnique (EWMOTE) for industrial fault diagnostics and prognostics. The methods comprises a weighted minority sampling strategy to identify hard-to-learn informative minority fault samples and Expectation Maximization (EM) based imputation algorithm to generate fault samples. To validate the performance of the proposed methods, experiments are conducted in two real datasets. The results show that the method could achieve better performance on not only binary class, but multi-class imbalance learning task in different imbalance ratios than other oversampling-based baseline models.

Motivation & Objective

• To address the challenge of class imbalance in data-driven fault diagnostics and prognostics for industrial systems, where minority fault samples are scarce and hard to learn.
• To improve the quality of synthetic minority samples by adapting to data distribution and identifying hard-to-learn informative samples.
• To develop a data-level oversampling technique that enhances classifier performance without relying on algorithm-level modifications.
• To validate the effectiveness of the proposed method on real-world industrial datasets with varying imbalance ratios.

Proposed method

• EWMOTE employs a weighted minority sampling strategy to prioritize hard-to-learn minority fault samples for synthetic generation.
• It uses the Expectation-Maximization (EM) algorithm to impute missing values and generate new minority-class samples based on the underlying distribution of existing minority samples.
• The method improves upon MWMOTE by integrating EM imputation to preserve data distribution and reduce noise in generated samples.
• It combines feature extraction without dimensionality reduction to maintain diagnostic information, followed by data-level oversampling before classification.
• The algorithm adaptively generates synthetic samples based on the statistical structure of minority classes, minimizing distortion of majority class boundaries.
• The framework is evaluated using a pipeline of feature extraction, data-level oversampling via EWMOTE, and classification, with performance assessed across multiple imbalance ratios.

Experimental results

Research questions

• RQ1 Can an adaptive oversampling method improve fault detection performance in industrial systems with highly imbalanced data?
• RQ2 How does EWMOTE compare to SMOTE, MWMOTE, and EMICIL in terms of precision, recall, and F-measure across different imbalance ratios?
• RQ3 Does the integration of EM-based imputation and weighted minority sampling lead to more effective synthetic samples than standard oversampling techniques?
• RQ4 How robust is EWMOTE in handling multi-class fault classification with varying degrees of class imbalance?
• RQ5 To what extent does EWMOTE reduce the generation of noisy or misleading minority samples compared to baseline methods?

Key findings

• On the wind turbine freezing failure dataset, EWMOTE achieved a recall of 0.8302, precision of 0.8573, and FAM of 0.8326, outperforming SMOTE, MWMOTE, and EMICIL.
• In the multi-fault plant dataset, EWMOTE achieved the highest F-measure (0.8793) for fault class F3 and maintained strong performance across all fault types.
• For high-imbalance classes like F5, EWMOTE showed a higher true positive rate and lower false positive rate than MWMOTE and EMICIL, respectively, indicating better robustness.
• EWMOTE improved F-measure by 1.8% over SMOTE and 0.7% over MWMOTE in the wind turbine dataset, demonstrating consistent gains in balanced performance metrics.
• The confusion matrix analysis confirmed that EWMOTE generated more accurate and less noisy samples, especially in high-imbalance scenarios, reducing misclassification of minority classes.
• EWMOTE maintained high precision (0.9536) and recall (0.9614) for normal samples, indicating minimal disruption to the majority class distribution.