In this paper, we are analyzing the use of advanced neural networks. We are focusing on the attention mechanisms and transformers to predict the probability of postoperative recurrence of patients with non-small lung cancer, creating both clinical and radiomic data. We are discovering that early recurrence is the main key to improvement for patients’ treatment outcomes who have gone through surgery before. Here, we have used a diverse dataset with clinical information, deep learning-based radiomic (DLR) data from CT scans, and handcrafted radiomic features (HCR). We have created a proper and efficient model by using advanced neural network design, automated tumor segmentation, and ensemble learning techniques. We have mixed attention-based neural networks and meta-learning techniques like stacking classifiers and transformers specifically adapted for medical imaging. We have achieved great accuracy in our ensemble model, which is over 75% and an AUC of 0.77, beating standard methods in recurrence prediction of lung cancer. In NSCLC recurrence prediction, these results show the importance of meta-learning methods for improving clinical judgment and advanced neural networks.