A Holistic Performance Comparison for Lung Cancer Classification Using Swarm Intelligence Techniques.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Microarray data for cancer classification is challenging due to high dimensionality, noise, and redundancy.
• Identifying informative genes is crucial for accurate cancer diagnosis and treatment strategies.

Purpose of the Study:

• To develop and evaluate a robust methodology for classifying high-dimensional lung cancer microarray data.
• To improve classification accuracy by integrating advanced feature selection and swarm intelligence optimization techniques.

Main Methods:

• Gene ranking using Information Gain, Relief-F, Chi-square, and T-statistic.
• Feature subset optimization via swarm intelligence algorithms (GO, MFO, BFO, KHO, AFSO).
• Classification using Naïve Bayesian, Decision Trees, SVM, and KNN classifiers.

Main Results:

• A hybrid approach combining Relief-F test for feature selection and Artificial Fish Swarm Optimization (AFSO) for optimization yielded optimal results.
• Using the top 100 genes identified by this hybrid method, Decision Trees achieved the highest classification accuracy of 99.10%.

Conclusions:

• The proposed two-stage methodology effectively addresses the challenges of high-dimensional microarray data in cancer classification.
• The combination of Relief-F, AFSO, and Decision Trees offers a highly accurate and promising approach for lung cancer diagnosis.