Cervical cancer is one of the most common and severe threats to women’s health. Early detection of abnormal cervical cells through automated screening can improve diagnostic accuracy, allowing for timely treatment and increased survival rates. This study proposes a solution that combines lightweight deep learning models with data augmentation of microscopic cell images to detect abnormal cervical cells. Lightweight models, including MobileNetV1, MobileNetV2, MobileNetV3 (both small and large variants), and EfficientNet-B0, were tested and demonstrated promising results after data augmentation. The EfficientNet-B0 model achieved 99% accuracy with an F1 score of 98%, while the MobileNet variants also showed high performance, with F1 scores ranging from 85% to 96% and a loss as low as 0.006. These experimental results highlight the potential of lightweight deep learning models combined with data augmentation to deliver high accuracy and efficiency, making them suitable for medical datasets with limited and imbalanced data across classes.