Abstract
BackgroundLung cancer remains a leading cause of cancer-related mortality, with accurate staging essential for guiding treatment. Advances in next-generation sequencing (NGS) and machine learning (ML) enable more precise classification, improving on traditional imaging-based methods.ObjectiveThis retrospective study applies XGBoost with cross-validation (CV) to classify early vs. late-stage lung cancer using RNA-Seq data from 993 patients in The Cancer Genome Atlas (TCGA) cohort.MethodsGene selection was conducted using the Wilcoxon rank-sum test on training data, and the XGBoost model was optimized via cross-validation. Model performance was assessed using the Area Under the Curve (AUC), with sensitivity-specificity analysis across classification thresholds.ResultsThe XGBoost model achieved a test AUC of 0.6534, identifying 40 key genes that optimize predictive accuracy while minimizing overfitting. Thresholds of 0.3 and 0.4 were optimal, balancing sensitivity and specificity for clinical application.ConclusionsIntegrating RNA-Seq data with machine learning improves lung cancer staging accuracy. Future research should focus on dataset expansion, model benchmarking, and multi-omics integration to enhance clinical applicability.