A hybrid ensemble method based on double disturbance for classifying microarray data.

Area of Science:

• Bioinformatics
• Machine Learning
• Computational Biology

Background:

• Microarray data presents challenges due to high dimensionality and the presence of irrelevant or redundant genes, hindering accurate classification.
• Existing classification methods may struggle with the complexity and noise inherent in microarray datasets.

Purpose of the Study:

• To propose a novel hybrid ensemble method designed to enhance classification performance on high-dimensional microarray data.
• To address the issues of irrelevant and redundant genes in microarray datasets through a multi-stage gene selection and reduction process.

Main Methods:

• A hybrid ensemble method incorporating a double disturbance strategy was developed.
• Gene ranking and selection using the reliefF algorithm, followed by bootstrap sampling to create diverse training subsets.
• Attribute reduction via neighborhood mutual information and ensemble construction using teaching-learning-based optimization with weighted voting.

Main Results:

• The proposed method demonstrated a reduction in ensemble size.
• Significantly higher classification performance was achieved compared to established methods like Bagging, AdaBoost, and Random Forest.
• Validation was performed on six benchmark cancer microarray datasets.

Conclusions:

• The hybrid ensemble method effectively handles the challenges of high-dimensional microarray data.
• The double disturbance approach improves classification accuracy and efficiency.
• This method offers a promising advancement for gene-based cancer classification.