Evolutionary extreme learning machine with sparse cost matrix for imbalanced learning.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Conventional extreme learning machines (ELMs) assume equal misclassification costs, limiting their effectiveness on imbalanced datasets.
• Skewed data distributions pose challenges for ELMs in accurately learning characteristics of few-shot cases.

Purpose of the Study:

• To develop a cost-sensitive learning approach for ELMs that addresses data imbalance.
• To enhance the representation of minority classes in machine learning models.

Main Methods:

• A case-weighting extreme learning machine (ELM) was developed using a sparse, diagonal cost matrix.
• A multi-objective optimization problem was formulated for penalty factors, solved using an evolutionary algorithm and an error bound model.
• The method implements adaptive cost-sensitive learning guided by generalization ability and case-weighting factors.

Main Results:

• The proposed method demonstrates competitive performance on benchmark and real-world datasets.
• Effective identification of software bug reports was achieved, even with imbalanced data distributions.
• Enhanced representation of few-shot cases was observed.

Conclusions:

• The developed case-weighting ELM effectively handles imbalanced data by incorporating adaptive cost-sensitivity.
• This approach offers a robust solution for classification tasks with skewed category distributions, particularly in software engineering applications.