A class imbalance-aware Relief algorithm for the classification of tumors using microarray gene expression data.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning in Oncology

Background:

• DNA microarray data is crucial for cancer research, enabling tumor classification.
• High-dimensional and imbalanced gene expression data present challenges for traditional machine learning classifiers.
• Existing Relief-based feature weighting methods often neglect class imbalance, potentially biasing results towards the majority class.

Purpose of the Study:

• To propose a novel method, imRelief, for efficiently handling high-dimensional imbalanced gene expression data.
• To correct the bias in feature weighting towards majority classes inherent in existing Relief algorithms.
• To improve the ability of feature weighting methods to identify features relevant to minority classes.

Main Methods:

• Development of the imRelief algorithm, a novel feature weighting technique.
• Modification of Relief-based methods to account for class imbalance and minority class sample distribution.
• Evaluation of imRelief on four distinct microarray gene expression datasets.

Main Results:

• imRelief effectively corrects bias towards the majority class in feature weight estimation.
• The proposed method considers the scattered distribution of minority class samples.
• Experiments demonstrate imRelief's effectiveness in both feature weighting and subset selection.

Conclusions:

• imRelief offers an improved approach for analyzing high-dimensional imbalanced gene expression data in cancer research.
• The method enhances the identification of discriminative features for minority classes.
• imRelief shows promise for building more robust and accurate cancer classifiers.