Ensemble Classification Model With CFS-IGWO-Based Feature Selection for Cancer Detection Using Microarray Data

Area of Science:

• Oncology
• Bioinformatics
• Machine Learning

Background:

• Cancer is a leading global cause of death, necessitating advancements in early detection.
• Machine learning (ML) offers promising approaches for early cancer diagnosis, utilizing gene expression and microarray data.
• High-dimensional data in ML models, common in gene expression and microarray datasets, can reduce efficiency.

Purpose of the Study:

• To propose and evaluate two ensemble techniques for improving ML-based cancer diagnosis.
• To investigate the effectiveness of Correlation Feature Selection (CFS) and Improved Grey Wolf Optimizer (IGWO) for feature selection and optimization.
• To compare the performance of majority voting and weighted average ensemble methods.

Main Methods:

• Utilized gene expression and microarray data for ML model training.
• Applied Correlation Feature Selection (CFS) for feature selection and Improved Grey Wolf Optimizer (IGWO) for feature optimization.
• Implemented ensemble techniques (majority voting and weighted average) to combine predictions from various classifiers including SVM, MLP, LR, DT, AdaBoost, ELM, and KNN.

Main Results:

• Evaluated model performance using Accuracy (ACC), Specificity (SPE), Sensitivity (SEN), Precision (PRE), Matthews Correlation Coefficient (MCC), and F1-score (F1-S).
• The majority voting ensemble technique demonstrated superior performance compared to the weighted average ensemble technique.
• Feature selection and optimization methods (CFS and IGWO) were crucial for handling high-dimensional data.

Conclusions:

• Ensemble methods, especially majority voting, significantly enhance the accuracy of ML-based cancer diagnosis.
• Effective feature selection and optimization are vital for managing high-dimensional omics data in cancer research.
• The proposed approach offers a robust strategy for early cancer detection, potentially reducing mortality rates.