A Multi-Dataset Stacked Ensemble Framework for Multi-Class Lung Cancer Classification

Abstract: Lung cancer is a major contributor to cancer-related deaths worldwide, emphasizing the need for accurate and early diagnostic systems. Although deep learning–based approaches have shown promising performance in classifying lung cancer from computed tomography (CT) images, many existing methods are limited by reliance on single datasets and restricted class categorization. To overcome these challenges, this paper introduces SEMLCC-5X, a stacked ensemble framework for multi-class lung cancer classification. The proposed approach integrates multi-source data by combining two publicly available CT datasets, improving data diversity and generalization. The classification task is extended to a fine-grained five-class problem, including normal, benign, adenocarcinoma, large-cell carcinoma, and squamous-cell carcinoma. The framework employs transfer learning with fine-tuning to train four deep learning base models by integrating custom CNN classification layers with the pre-trained Xception, VGG19, EfficientNetB7, and InceptionV3 architectures. These models are combined using a stacking strategy with a logistic regression meta-learner for optimized prediction aggregation. The SEMLCC-5X model is evaluated using comprehensive metrics, including accuracy, precision, recall, F1-score, and AUC. Experimental results demonstrate superior performance, achieving an accuracy of 98.06% and a macro AUC of 99.58%, outperforming individual models and conventional ensemble methods. In conclusion, SEMLCC-5X provides a robust and accurate framework for multi-class lung cancer classification, with strong potential for integration into computer-aided diagnostic systems and clinical workflows.

Keywords: Medical Image Analysis, Computed Tomography (CT), Stacked Ensemble Learning, Deep Learning, Multi- Class Lung Cancer Classification.