Optimizing cancer classification: a hybrid RDO-XGBoost approach for feature selection and predictive insights

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• High-dimensional cancer data presents challenges for biomarker identification due to complexity and heterogeneity.
• Conventional feature selection methods often lack efficiency and predictive accuracy in complex cancer datasets.

Purpose of the Study:

• To develop a novel feature selection framework integrating Random Drift Optimization (RDO) with XGBoost for enhanced cancer classification.
• To improve the identification of relevant biomarkers and gain insights into cancer progression mechanisms.

Main Methods:

• Integration of Random Drift Optimization (RDO) with XGBoost algorithm for feature selection.
• Application and validation of the proposed framework on diverse real-world cancer datasets (CNS, Leukemia, Breast, Ovarian).
• Comparative analysis against Support Vector Machine, K-Nearest Neighbor, and Naive Bayes classifiers.

Main Results:

• The RDO-XGBoost framework significantly improved classification accuracy and efficiency across multiple cancer types.
• Successfully identified a smaller subset of unique and relevant genes, aiding in understanding cancer biology.
• Achieved high accuracy rates: 97.24% (CNS), 99.14% (Leukemia), 95.21% (Ovarian), and 87.62% (Breast cancer).

Conclusions:

• The RDO-XGBoost framework demonstrates superior performance compared to traditional classifiers for cancer data analysis.
• The method offers a promising solution for robust feature selection, enhanced predictive performance, and biological insights in cancer research.