Cancer Classification Utilizing Voting Classifier with Ensemble Feature Selection Method and Transcriptomic Data

Area of Science:

• Bioinformatics
• Machine Learning
• Cancer Genomics

Background:

• High-dimensional gene expression data presents challenges for accurate cancer diagnosis.
• Existing feature selection algorithms struggle with identifying critical genes in complex datasets.

Purpose of the Study:

• To develop an effective ensemble rank-based feature selection method (EFSM) for identifying important genes.
• To create an ensemble weighted average voting classifier (VT) for improved cancer classification.
• To enhance the accuracy and stability of machine learning models in cancer identification.

Main Methods:

• Proposed an ensemble rank-based feature selection method (EFSM) aggregating features from multiple methods.
• Developed an ensemble weighted average voting classifier (VT) combining Support Vector Machine, k-Nearest Neighbor, and Decision Tree algorithms.
• Validated the proposed method on three benchmark cancer datasets.

Main Results:

• Achieved high classification accuracy: 100% for leukemia, 94.74% for colon cancer, and 94.34% for an 11-tumor dataset.
• Identified a subset of crucial cancer-related genes with demonstrated significance.
• The proposed approach outperformed existing ensemble models in accuracy and stability.

Conclusions:

• The EFSM and VT provide a robust and accurate approach for biomarker-based cancer identification.
• The identified key genes are vital for improving machine learning-based gene analysis in oncology.
• This study significantly advances the field of machine learning applications in cancer genomics.