A Lightweight and Explainable Machine Learning Framework for Cervical Cancer Risk Prediction

Department of Computer Science & Engineering, Brainware University, Barasat, Kolkata

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Cervical cancer is a global health issue of serious concern especially in the low resource areas where access to good screening facilities is a hindrance. Although the device of machine learning has been broadly examined in regard to cancer prediction, numerous of the current methods apply deep learning frameworks that demand a substantial amount of computing resources and are not interpretable, which limits their practical usage. The present paper gives a lightweight and interpretable model of machine learning-based cervical cancer risk prediction based on structured clinical and demographic data. The suggested structure compares several conventional machine learning models such as the Logistic Regression, Support Vector Machine, Decision Tree, and the Random Forest. The Synthetic Minority Over-sampling Technique (SMOTE) is used to address the problem of class imbalance, whereas the problem of missing values is addressed with the help of median-based imputation. Stratified cross-validation is used k-foldly to make sure that the evaluation of performance is healthy and not biased. These models are evaluated based on clinically relevant measures such as accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (ROC-AUC) and analysis of computational efficiency. Through experiment, it has been established that the Random Forest classifier is more successful with the accuracy of over 96, high recall rates of cancer-positive cases, and high robustness when features are missing. SHAP-based feature attribution is used to incorporate model explainability, which gives clear information about the factors that affect the risk. The consistency of the results obtained in cross-validation folds stability analysis is another indicator of the efficiency of the suggested method. The findings suggest that interpretable and lightweight machine learning models can be used to provide the correct and deployable cervical cancer screening solutions that can fit in healthcare settings with a limited number of resources.