Performance Evaluation of Machine Learning Models For Cervical Cancer Prediction

Abstract

Cervical cancer is exclusively an anatomy of the female genitals involving the cervix and is the common cancer type that appears in all age women groups and the most common cause of death associated with cancer in gynecological practice, yet it is almost completely preventable if precancerous lesions are identified and treated promptly. The need to develop a quick, cheap and efficient method to diagnose a precursor lesion in an environment with high burden of the diseases with a view of reducing the burden of the disease motivated the need to apply Machine Learning (ML) technique towards cancer prediction. The primary objective of the study was to develop a ML model that can predict the occurrence of cervical cancer with a higher degree of accuracy. The cervical cancer dataset used in this study was obtained from Jos University Teaching Hospital (JUTH) and Aids Prevention Initiative in Nigeria (APIN). Several ML techniques were considered which includes Ensemble Bagged Tree, Fine Gaussian SVM, Cubic SVM, Fine Tree, Quadratic SVM, Medium Gaussian SVM, Ensemble Boosted Tree, Ensemble Rusboosted Tree, Medium Tree, Linear SVM, Corase Gaussian SVM and Coarse Tree algorithm. The study shows that Ensemble Bagged Tree and Fine Gaussian SVM gives a higher cervical cancer predictive accuracy of 99.7 percent and 99.6 percent respectively as the best performing predictive models, followed by Cubic SVM and Fine Tree with 98 percent and Fine Tree with 96.6 percent cervical cancer predictive accuracy respectively. The performance evaluation shows that Ensemble Bagged Tree and Fine Gaussian SVM perform excellently well in distinguishing and predicting the cervical classes correctly with the best prediction accuracy.