Enhancing Cancerous Gene Selection and Classification for High-Dimensional Microarray Data Using a Novel Hybrid...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Microarray datasets are crucial for early cancer diagnosis but often contain noisy and irrelevant genes.
• High dimensionality of microarray data poses challenges for machine learning algorithms.
• Effective feature selection is vital for accurate cancer classification.

Purpose of the Study:

• To develop a hybrid feature selection method for high-dimensional microarray datasets.
• To improve cancer diagnosis and classification accuracy.
• To identify the most influential genes for cancer prediction.

Main Methods:

• A two-phase hybrid feature selection model combining filter methods and differential evolution (DE) optimization.
• Selection of top-ranked features using popular filter methods.
• Further optimization of feature selection using DE algorithm.
• Training machine learning models on selected optimal features for cancer classification.

Main Results:

• Achieved classification accuracies of 100% (Brain, CNS), 93% (Breast), and 98% (Lung).
• DE-based feature selection reduced features by approximately 50% compared to filter methods alone.
• Demonstrated significant average accuracy improvements over previous works, up to 57.45%.

Conclusions:

• The proposed hybrid filter-DE method shows superior performance in improving cancer classification accuracy.
• The method effectively reduces feature dimensionality while retaining crucial information.
• The approach offers a promising strategy for enhanced cancer diagnosis using genomic data.