Advanced Cancer Classification Using AI and Pattern Recognition Techniques

Abstract

Accurate cancer classification is essential for early detection and effective treatment, yet the complexity of gene expression presents significant challenges. In this study, we explored how combining multiple feature selection methods with various classifiers enhances the identification of marker genes for four cancers: leukemia, lung, lymphoma, and ovarian cancer. We applied feature selection techniques such as the F Test, Signal-to-Noise Ratio (SNR), T-test, ReliefF, Correlation Coefficient, Mutual Information, and minimum redundancy maximum relevance, along with classifiers including K-Nearest Neighbors, Support Vector Machines, Linear Discriminant Analysis, Decision Tree Classifiers, and Naive Bayes. Our results demonstrate that the SNR method consistently achieved the highest accuracy in gene selection, particularly when paired with K-means clustering. Remarkably, leukemia was classified with 100% accuracy using only four genes, lung cancer, and lymphoma with 100% and 97% accuracy, respectively, using three genes, and ovarian cancer with 100% accuracy using just one gene. These findings highlight the potential of minimal gene sets for highly precise cancer classification.