Robustification of Naïve Bayes Classifier and Its Application for Microarray Gene Expression Data Analysis.

Area of Science:

• Bioinformatics
• Machine Learning
• Genomics

Background:

• Microarray gene expression data (MGED) analysis commonly uses the naïve Bayes classifier (NBC).
• Classical NBC is sensitive to outliers, a frequent issue in MGED due to experimental processes.
• Outliers can significantly impair the accuracy of gene expression data classification.

Purpose of the Study:

• To develop a robust Gaussian naïve Bayes classifier (NBC) for microarray gene expression data (MGED).
• To address the sensitivity of classical NBC to outliers using a minimum β-divergence method.
• To evaluate the performance of the proposed β-NBC against existing classifiers.

Main Methods:

• Robustification of the Gaussian NBC using the minimum β-divergence method.
• Estimation of robust location and scale parameters from training data.
• Outlier detection and modification in test data using the β-divergence parameter.
• Comparative analysis with NBC, KNN, SVM, and AdaBoost on simulated and real gene expression datasets.

Main Results:

• The proposed beta naïve Bayes classifier (β-NBC) demonstrated improved performance in the presence of outliers.
• The β-NBC achieved comparable performance to traditional methods when datasets were not contaminated by outliers.
• The minimum β-divergence method effectively produced robust estimators for location and scale parameters.

Conclusions:

• The β-NBC offers a robust alternative for gene expression data analysis, particularly when outliers are present.
• The minimum β-divergence method successfully enhances the performance of NBC in noisy biological datasets.
• The proposed method provides a valuable tool for accurate pattern recognition in MGED.