A Deep-Ensemble-Level-Based Interpretable Takagi-Sugeno-Kang Fuzzy Classifier for Imbalanced Data.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Imbalanced data classification suffers from high misclassification rates due to problematic areas like rare samples and class overlap.
• Existing methods often struggle with the complexities introduced by these challenging data characteristics.

Purpose of the Study:

• To present a novel deep ensemble-level Takagi-Sugeno-Kang fuzzy classifier (IDE-TSK-FC) for imbalanced data classification.
• To achieve both high classification performance and interpretability using fuzzy classifiers.
• To enhance generalization capability in class imbalance learning by elevating oversampling to the deep ensemble level.

Main Methods:

• A deep ensemble of zero-order Takagi-Sugeno-Kang (TSK) fuzzy subclassifiers is stacked on minority class problematic areas.
• Successive subclassifiers are built layer-by-layer on newly identified problematic areas and averaged predictions from previous layers.
• K-nearest neighbors identify problematic areas, incorporating random feature/membership function selection and a least learning machine for rule consequents.

Main Results:

• The proposed IDE-TSK-FC demonstrates superior performance in class imbalanced learning across public and real-world healthcare datasets.
• The method effectively handles problematic data areas, leading to reduced misclassification rates.
• Stacking fuzzy subclassifiers at the ensemble level improves generalization capability.

Conclusions:

• IDE-TSK-FC offers a promising approach for imbalanced data classification, balancing performance and interpretability.
• The deep ensemble strategy effectively addresses challenges posed by imbalanced datasets.
• This novel method shows significant potential for applications in various domains, including healthcare.