With the rapid increase in population, the rate of diseases like cancer is also increasing. Lung cancer is a leading cause of cancer-related deaths with a minimum survival rate; there is a need to find better, faster, and more accurate methods for early diagnosis of this disease. Although previous research in lung cancer has presented numerous prediction schemes, the feature selection utilized in the schemes and learning process has failed to enhance the accurate performance of lung cancer diagnosis, including incorrect classification and low prediction levels, which lead to misdiagnosis. Prediction of lung cancer cells from lung images in early stages is a question mark for researchers. This study presents a discerning way of predicting lung cancer with the Grey Wolf Optimization Algorithm (GWOA) and Convolutional Neural Networks (CNN). The 14,740 CT scan images are used for classification. The Kaggle dataset, data preprocessing, hyper-parameter feature selection using GWOA, classification using CNN, RF, and DT, cross-validation, and classifier evaluation are the five phases of the proposed lung cancer prediction architecture. The noise present in the data was eliminated by applying a bin smoothing normalization process. In terms of lung cancer prediction, we show that the highest score is achieved when applying CNN with GWOA, which produced the best results with an average performance of 96% accuracy, F1-score, precision, and recall, respectively compared to RF and DT with GWOA. Similarly, the CNN-GWOA produced the lowest false negative rate (FNR) of 0.023676. The low FNR means that it was possible to diagnose lung cancer with very minimal incorrect classification errors. This translates to successful prediction of lung cancer disease correctly.