Cancer Classification with a Cost-Sensitive Naive Bayes Stacking Ensemble.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning in Medicine

Background:

• Ensemble learning enhances model flexibility and generalization.
• Cancer gene expression data often presents imbalanced characteristics, posing classification challenges.
• Effective feature selection is crucial for accurate cancer classification.

Purpose of the Study:

• To develop a robust ensemble learning method for high-quality cancer classification.
• To address the challenge of imbalanced cancer gene expression data.
• To evaluate the proposed method's performance against existing classification techniques.

Main Methods:

• Employed fast correlation-based feature selection (FCBF) for data preprocessing.
• Utilized a stacking ensemble learner with Support Vector Machine (LIBSVM), K-nearest neighbor (KNN), C4.5 decision tree, and Random Forest (RF) as base learners.
• Integrated a cost-sensitive naive Bayes (CSNB) as the meta-learner, termed CSNB stacking, to handle data imbalance.

Main Results:

• The CSNB stacking method demonstrated effectiveness and robustness across nine diverse cancer datasets.
• Experimental results confirmed superior performance compared to single classifiers and other ensemble algorithms.
• The method successfully processed imbalanced cancer gene expression data, yielding accurate classifications.

Conclusions:

• The proposed CSNB stacking method offers a promising approach for accurate cancer classification.
• This technique can potentially aid in cancer diagnosis and advance cancer research.
• The study highlights the value of ensemble learning and cost-sensitive methods in bioinformatics.