PREDICTIVE MODELING FOR EARLY LUNG CANCER DETECTION USING CTGAN-AUGMENTED FEATURES AND TREE-BASED LEARNING TECHNIQUES

Abstract

Early detection of lung cancer significantly improves patient survival rates by enabling timely clinical intervention. However, the development of accurate predictive models is challenged by limited availability of labeled medical datasets, class imbalance, and the complexity of radiological features extracted from computed tomography (CT) images. This research proposes a predictive modeling framework that integrates Conditional Tabular Generative Adversarial Network (CTGAN)-based data augmentation with advanced tree-based machine learning techniques for early lung cancer detection. The proposed approach utilizes CT-derived clinical and morphological features, enhances minority class representation through synthetic feature generation, and applies ensemble learning algorithms including Random Forest, Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and CatBoost. The effectiveness of CTGAN augmentation is evaluated by comparing model performance with and without synthetic data enhancement. Experimental analysis demonstrates that CTGAN-generated samples improve model generalization, reduce classification bias, and enhance predictive accuracy. The proposed framework provides an efficient and interpretable solution for computer-aided lung cancer diagnosis and supports early-stage clinical decision-making.