Cervical cancer is a significant threat to women’s health, ranking as the fourth most common cancer among women worldwide. Globally, the Globocan 2020 initiative has detected 6,40,000 new cases of cervical cancer, which has resulted in 3,42,000 cancer-related fatalities. One of the primary methods for early detection of cervical cancer is the Pap smear test, which involves examining cervical cells under a microscope to identify abnormalities. However, this approach is labor-intensive, subjective, time-consuming and prone to human error. Recent advancements in deep learning have increased the precision and effectiveness of Pap smear diagnosis thereby reducing the workload of medical experts in this domain. In this paper, we propose a novel approach utilizing a custom EfficientNet model, which is pre-trained on widely used MNIST-digit dataset and is then applied on the public cervical cancer datasets available, namely Liquid Based Cytology (LBC) dataset and SIPAKMED dataset. Our approach aims to classify smear images as either normal or cancerous, offering a more automated and reliable solution for cervical cancer screening. This proposed model has achieved an accuracy of 97.14% for LBC dataset whereas 84.30% for SIPAKMED dataset respectively.