OPTIMIZING VGG-16 CONVOLUTIONAL NEURAL NETWORK FOR PAP SMEAR IMAGE CLASSIFICATION IN CERVICAL CANCER DETECTION

Early detection of cervical cancer through Pap smear image analysis plays a crucial role in reducing mortality rates associated with this disease. This study aims to optimize the VGG16 architecture to improve the classification accuracy of Pap smear images. The proposed method employs transfer learning with pre-trained ImageNet weights, customization of fully connected layers, and data augmentation techniques to enhance the diversity of training images. Experimental results demonstrate a significant improvement in training accuracy, reaching 98.50%, while validation accuracy remained stable at 88.24%, indicating potential overfitting. Performance testing on unseen data yielded an accuracy of 80%, with high precision for the negative class but low recall for the positive class, suggesting a bias toward the majority class. These findings highlight the need for additional strategies, such as data balancing and hybrid method integration, to improve sensitivity to positive cases. This research contributes to the development of adaptive deep learning-based classification models that support clinical decision-making in cervical cancer screening and opens opportunities for further research on model optimization and dataset expansion.