Enhancing Lung Cancer Type Prediction with a Novel Hybrid Approach: Transfer Learning, SVM, and Model Stacking

In this paper, we propose a novel hybrid model for the prediction of multiclass lung cancer, an ailment of paramount concern in the medical field. We leverage transfer learning applied to Convolutional Neural Networks (CNN), alongside Support Vector Machines (SVM), and Principal Component Analysis (PCA), combined via stacking to form a powerful meta-model. The model utilizes pre-trained CNN architectures to extract intricate features from medical imaging data, which then undergo dimensionality reduction via PCA, enhancing the computational efficiency while preserving vital information. Following feature extraction and reduction, we introduce an SVM, known for its exceptional classification prowess, to construct a preliminary predictive model. To maximize the performance, we employ a stacking methodology, treating the preliminary models as base classifiers and training a meta-learner on their predictions to predict the final class labels. Experimental results on real-world lung cancer datasets demonstrate that our model significantly outperforms traditional machine learning and deep learning models in terms of prediction accuracy, sensitivity, and specificity. This work contributes to enhancing the accuracy of multiclass lung cancer predictions, potentially saving lives by facilitating early and accurate diagnoses. Further research will focus on applying this model to other forms of cancer and diseases.