Cervical Cancer Prediction Using Machine Learning with Feature Selection and Performance Evaluation

Abstract

Introduction: The issue of cervical cancer has been one of the most eminent health issues of women all over the world, particularly in the developing world where early screening and diagnosis is not very prevalent. The early diagnosis is significant in improving the survival and mortality.
Objectives: This study introduces an in-depth machine learning-based model of cervical cancer prediction based on clinical, behavioral, and demographic information.
Methods: The dataset utilized in the research is publicly available and includes 835 records of patients and 34 features, and the variable of interest was the biopsy results. This data is highly pre-processed: it is dealing with missing values, Synthetic Minority Over-sampling Technique (SMOTE) to correct the problem of a class imbalance, and Interquartile Range (IQR) to eliminate outliers. Different machine learning and deep learning algorithms like Random Forest, K-Nearest Neighbors (KNN), XGBoost, Artificial Neural Networks (ANN), and Deep Neural Networks (DNN) are used and contrasted. To improve the interpretability and efficiency of models, the process of feature selection is carried out with the help of permutation importance.
Results: The experimental findings show that the Random Forest model with the selected features has the best accuracy of 97.74% which is higher than other models with being robust and interpretable.
Conclusions: This paper demonstrates the usefulness of machine learning algorithms to predict cervical cancer at its initial stages and how it may be used to enhance healthcare decision-making.