Learning misclassification costs for imbalanced classification on gene expression data.

Area of Science:

• Bioinformatics
• Machine Learning
• Computational Biology

Background:

• Cost-sensitive learning effectively addresses imbalanced classification challenges.
• Empirical determination of misclassification costs leads to performance instability.
• An efficient and accurate method for optimal cost weight calculation is crucial.

Purpose of the Study:

• To propose and compare two novel approaches for calculating optimal cost weights.
• To enhance the performance of cost-sensitive classification algorithms.
• To achieve the highest weighted classification accuracy (WCA) in imbalanced datasets.

Main Methods:

• Developed two distinct methods: optimal cost weights grid searching and function fitting.
• Employed extreme learning machine (ELM) for classifying imbalanced gene expression data.
• Compared the efficiency and effectiveness of the proposed cost weight calculation approaches.

Main Results:

• Both grid searching and function fitting were evaluated for optimal cost weight determination.
• Function fitting demonstrated superior efficiency in identifying optimal cost weights.
• The proposed methods were tested on imbalanced gene expression datasets.

Conclusions:

• The function fitting method is generally more efficient for finding optimal cost weights.
• This approach achieves acceptable weighted classification accuracy (WCA).
• The findings suggest function fitting as a preferred strategy for cost-sensitive learning.