A Comparative Study of Data Mining Methods to Diagnose Cervical Cancer

Abstract

Cervical cancer becomes a major cause of cancer deaths in women around the world. The objective of this study is to provide a comprehensive analysis of different data mining methods to diagnose the malignant cancer samples. Different data mining algorithms (SVM, Naïve Bayes, and KNN) has been applied on four different medical tests (Biopsy, Cytology, Hinselmann, and Schiller) as four different target variables. The attributes influence the disease most is extracted since the disease has no symptoms in the early stage. The extraction involved over 32 attributes and two different algorithms such as Correlation-based Filter (CFS) and Random Forest. The results showed that the performance of Naïve Bayes classifier outperforms other classifiers after evaluation using 10-fold cross-validation method in R environment. In addition, the use of attribute selection has been proved not only can select the highly important attributes but also to increase the performance of all classifiers on cervical cancer dataset. In this study, the work reveals the classifiers can effectively achieve the best performance with the least number of highly important attributes.