Class-specific extreme learning machine for handling binary class imbalance problem.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Class imbalance, where majority class instances outnumber minority ones, biases conventional extreme learning machine (ELM) predictions.
• Existing ELM variants like Weighted ELM and class-specific cost regulation ELM (CCR-ELM) attempt to address this imbalance.
• These methods often involve complex weighting schemes or do not fully account for class distribution and overlap.

Purpose of the Study:

• To introduce a novel variant of the extreme learning machine, termed class-specific extreme learning machine (CS-ELM).
• To effectively handle binary class imbalance problems with improved accuracy and reduced computational load.
• To propose a method that utilizes class-specific regularization parameters computed based on class distribution.

Main Methods:

• The proposed CS-ELM method modifies the standard ELM architecture for imbalanced datasets.
• It employs class-specific regularization parameters derived from class distribution, differing from CCR-ELM.
• The computation of the output weight (β) is also distinct from existing approaches, aiming for efficiency.

Main Results:

• CS-ELM demonstrated superior performance compared to Weighted ELM, CCR-ELM, EFSVM, FSVM, and SVM on benchmark imbalanced datasets.
• The proposed method effectively mitigates bias towards the majority class inherent in conventional ELM.
• CS-ELM exhibits lower computational overhead than both Weighted ELM and CCR-ELM.

Conclusions:

• CS-ELM is a highly effective approach for tackling binary class imbalance problems in machine learning.
• The method offers a computationally efficient alternative with enhanced predictive accuracy over existing ELM variants.
• CS-ELM's performance on real-world imbalanced datasets validates its practical applicability and superiority.