A Novel Weighted Ensemble Strategy Using Transfer Learning for Enhanced Cervical Cancer Classification

Abstract

Cervical cancer (CC) has become one of the major concerns for the researchers. Early detection, along with proper medication and treatment, can reduce the prevalence of this life-threatening cancer. In the burgeoning realm of Artificial Intelligence (AI), CC detection has seen a paradigm shift towards AI-driven early diagnosis systems. Through various studies employing diverse datasets and methodologies, researchers are advancing the frontier of cervical cancer detection, harnessing the potential of AI for improved accuracy and efficiency. The previous study has limitations such as accuracy, issues, and a small dataset. The present work introduces a novel weighted ensemble model using multiple pre-trained CNN models. The ensemble weights are calculated using a unique strategy, optimizing the combined predictive power of the models. Furthermore, the thorough data augmentation methods improve the suggested model's resilience and dependability. To overcome the dataset limitations, we have applied data augmentation techniques in this research to increase the data samples artificially. We have achieved 99.65% accuracy in this research and outperformed the previous model. The precision, recall, and f-score are accordingly 99%. This study lays a new genre for further research and creates a new scope for researchers in medical imaging and diagnostics.