AI-Driven Cervical Cancer Detection: Autoencoder-Enhanced Model Optimization and Interpretability
V. Gandhiraj, Arunkumar RS, M. Rajalakshmi, C. Parthasarathy, T. Dinesh Kumar, Manne Archana · 2025
This research work examines the amalgamation of autoencoder-based dimensionality reduction with a Random Forest classifier to improve the early diagnosis of cervical cancer. The proposed methodology utilizes a real-world cervical cancer dataset to condense high-dimensional input features into latent representations, therefore considerably diminishing data complexity while preserving critical patterns. The compressed characteristics are utilized to train a Random Forest classifier, yielding enhanced predicted accuracy. The method attained competitive performance measures, including an AUC-ROC, demonstrating its efficacy in differentiating cervical cancer cases. The study emphasizes the interpretability of model predictions by utilizing feature significance analysis to pinpoint key elements affecting classification results. This research highlights the efficacy of integrating modern machine learning methodologies with interpretable frameworks, providing a robust and elucidative approach for enhancing medical data classification and facilitating clinical decision-making.