An Explainable SMOTE-Enhanced Fuzzy Ensemble Decision Tree Framework for Intelligent Cancer Prediction and Classification with Hyperparameter Turning

Abstract

Early detection of cervical cancer remains a critical challenge in healthcare, particularly in low- and middle-income countries where limited access to screening and diagnostic facilities contributes to high mortality rates. Although machine learning techniques have shown considerable promise in supporting cancer diagnosis, their effectiveness is often hindered by class imbalance, data uncertainty, and limited model interpretability. This study proposes an explainable SMOTE-enhanced fuzzy ensemble decision tree framework with hyperparameter optimization for intelligent cervical cancer prediction and classification. The framework integrates Synthetic Minority Oversampling Technique (SMOTE), fuzzy logic, multiple fuzzy decision trees (Fuzzy ID3, Fuzzy C4.5, and Fuzzy CART), and ensemble learning models, including Random Forest, Gradient Boosting, XGBoost, LightGBM, and AdaBoost, to improve predictive accuracy, robustness, and clinical interpretability. The Cervical Cancer (Risk Factors) dataset obtained from the UCI Machine Learning Repository, comprising 858 patient records and 36 clinical and demographic attributes, was employed for experimentation. Data preprocessing involved missing-value imputation, normalization, fuzzification of selected clinical variables using triangular membership functions, and class balancing through SMOTE. Hyperparameter tuning was performed using Bayesian optimization, grid search, and randomized search techniques. Model performance was evaluated using accuracy, precision, recall, F1-score, ROC-AUC, confusion matrices, and ROC curves. Experimental results demonstrated that ensemble methods consistently outperformed conventional decision tree algorithms. The proposed framework achieved its best performance with Gradient Boosting, attaining an accuracy of 97.09%, precision of 87.50%, recall of 63.64%, F1-score of 73.68%, and a ROC-AUC of 94.35%, while XGBoost and Random Forest achieved the highest discriminative capabilities with ROC-AUC values of 97.35% and 97.21%, respectively. The incorporation of SMOTE significantly improved minority-class detection, whereas fuzzy logic enhanced the handling of uncertainty inherent in medical data. The findings confirm that the proposed explainable framework provides an effective, robust, and interpretable decision-support mechanism for early cervical cancer diagnosis and has strong potential for broader applications in intelligent healthcare systems.