A precise machine learning model: Detecting cervical cancer using feature selection and explainable AI.

Area of Science:

• Oncology
• Medical Informatics
• Artificial Intelligence

Background:

• Cervical cancer presents a significant global health challenge, exacerbated by screening disparities and leading to increased mortality.
• Early detection and effective management are crucial for improving patient outcomes and reducing the disease burden.

Purpose of the Study:

• To develop and evaluate machine learning models for the early prediction of cervical cancer.
• To identify key risk factors and enhance diagnostic frameworks using advanced computational techniques.

Main Methods:

• Six machine learning models (Decision Tree, Logistic Regression, Naïve Bayes, Random Forest, K-Nearest Neighbors, Support Vector Machine) were applied to analyze 36 risk factors in 858 individuals.
• Data imbalance was addressed using Synthetic Minority Oversampling Technique (SMOTE) and Adaptive Synthetic Sampling (ADASYN).
• Feature selection was performed using Chi-square and LASSO, with model interpretability enhanced by Shapley Additive Explanations (SHAP).

Main Results:

• The Decision Tree (DT) model demonstrated superior performance, achieving 97.60% accuracy, 98.73% sensitivity, 80% specificity, and 98.73% precision with Chi-square feature selection.
• Even with data imbalance, the DT model maintained high performance with 97% accuracy, 99.35% sensitivity, 69.23% specificity, and 97.45% precision.
• Model performance was rigorously evaluated using metrics like accuracy, sensitivity, specificity, precision, F1-score, FPR, FNR, and AUC.

Conclusions:

• Machine learning models, particularly Decision Trees, show significant potential for accurate early-stage cervical cancer prediction.
• The integration of feature selection and explainable AI enhances the reliability and clinical applicability of these diagnostic tools.
• This research contributes to developing automated diagnostic frameworks for improved cervical cancer detection, management, and personalized patient care.