Lung Cancer Classification Using Stacking Framework of BiLSTM, Logistic Regression, and XGBoost

Abstract

Lung cancer remains one of the most prevalent and deadly cancers worldwide, causing over 1.8 million deaths each year. Early and accurate classification of lung cancer is crucial, yet existing machine learning and deep learning models often face limitations in generalization and reliability. To address this issue, this study proposes a stacking framework that integrates Bidirectional Long Short-Term Memory (BiLSTM) and Logistic Regression as base learners, with Extreme Gradient Boosting (XGBoost) serving as the meta-learner. The rationale for this approach is that BiLSTM captures complex feature interactions, Logistic Regression provides interpretability, and XGBoost has demonstrated strong performance as a meta-learner in ensemble studies. The framework was evaluated on a publicly available lung cancer dataset consisting of 309 patient records with 15 clinical and lifestyle attributes. Experimental results showed that the stacking framework achieved perfect accuracy of 1.00, outperforming BiLSTM (0.95) and Logistic Regression (0.93). These findings confirm the effectiveness of the proposed ensemble in overcoming the weaknesses of individual models and highlight its novelty as a reliable approach for lung cancer classification.