The most common cause of cancer-related deaths worldwide is lung cancer. An early and accurate diagnosis is required. Traditional methods used to diagnose lung cancer are X-rays and manual computed tomography (CT) analysis. The task is slow and has a high likelihood of errors because it requires manual efforts and is time consuming. The proposed work explores the use of deep learning techniques to classify lung CT scans into benign, malignant, and normal categories, using advanced models designed for efficient and accurate performance. The proposed work uses the IQ-OTHNCCD dataset, sourced from Kaggle, which consists of 1097 CT images classified into benign (120), malignant (561), and normal (416) categories. This dataset plays a crucial role in lung cancer diagnostics by providing various CT images, enabling models to effectively learn and differentiate between various lung conditions. Models including EfficientNetV2, MobileNetV3, and MobileNetV4 were employed to evaluate their effectiveness in this proposed work. The class imbalance was addressed using data augmentation techniques, which improved model dependability. Among the models tested, EfficientNetV2 achieved an accuracy of 98.82%, making it suitable for systems that require high computational capacity. MobileNetV4, with an accuracy of 98.53%, is better suited for deployment on edge devices due to its lightweight architecture and lower hardware requirements. The purpose of the proposed work is to increase the precision and effectiveness of lung cancer classification on CT scans, giving physicians a reliable option.