Lung cancer classification is one of the biggest challenges in modern medicine, and it is one of the deadliest types of cancer in the world. Building accurate classification models that can accurately predict different lung cancer types can be particularly challenging with a limited amount of data, irrelevant features, and imbalanced distribution in the datasets. To address this challenge, this study utilizes a two-stage framework where the data is balanced using the synthetic minority oversampling technique (SMOTE), and the second stage uses the preprocessed dataset on three machine learning (ML) techniques comprising of Decision Tree (DT), Extreme Gradient Boosting (XGB), and Logistic Regression (LR) with a soft voting technique. The SMOTE technique plays a significant role in improving the performance of all the models by balancing the minority class and mitigating the bias introduced by the class imbalance. Across all the experiments, the proposed model achieves an accuracy of 95.16%, precision of 95.04, and AUC score of 0.98, outperforming the results of all the individual models and all benchmarks taken for this study.