Performance Assessment of Machine Learning Classifiers Using Selective Feature Approaches for Cervical Cancer Detection

Worldwide, cervical cancer is the leading cause of death among women from cancer. The symptoms of this gynecological disease are difficult to recognize at early stage, especially in those countries that don't have facility of screening programs. In diagnosis of cervical cancer, machine learning methods can be used to detect the malignous cancer cells at initial stage. The foremost apprehension in disease diagnosis involves data imbalance issue and non-uniform scaling in dataset. In this article, a prevalent oversampling approach Synthetic Minority Oversampling Technique along with fivefold cross-validation is being used on unscaled and scaled data to handle these issues. A promising comparison is been made among the performance of most prevalent machine learning (ML) classifiers such as Naive Bayes, Logistic Regression, K-Nearest Neighbor, Support Vector Machine (SVM), Linear Discriminant analysis, Multi-Layer Perceptron, Decision Tree (DT) and Random Forest (RF) on unscaled data and scaled data obtained by Min-Max scaling, Standard scaling and Normalization. RF, SVM and DT are the top three ML algorithms obtained in cervical cancer diagnosis for which optimization possibilities are explored with feature selection methods as Univariate feature selection and Recursive feature elimination (RFE). Overall performance of Random Forest predictor with RFE (RF-RFE) is superior to all others being implemented.

Automated summary

This article discusses the use of machine learning methods for cervical cancer diagnosis, addressing the issues of data imbalance and non-uniform scaling. By comparing the performance of different machine learning algorithms under different data processing methods, it is found that the random forest predictor performs the best when optimized with feature selection methods.