Comparison of Performances of Associative Classification Methods for Cervical Cancer Prediction: Observational Study
Fatma Hilal Yağın, Burak Yagin, Ahmet Kadir Arslan, Cemil Çolak
Abstract
Objective: Associative classification is a method that generates a rule-based classifier in a categorical data set. The main purpose of the associative classification is to create classification models with high performance and, in addition, to improve interpretability thanks to the rules it creates. In this study, it is aimed to classify, predict cervical cancer with the methods of relational classification and to determine the most important parameters and relational rules associated with the disease. Material and Methods: In the study, regular class association rules (RCAR) and classification based on associations (CBA) methods were applied to the open access data set named 'Cervical Cancer Behavioral Risk Data Set' and the results were compared. In order to separate the numerical variables in the data set, Boruta feature selection method was applied to determine the most important features about Ameva and cervical cancer. The performances of the created relational classification models were evaluated with accuracy, balanced accuracy, sensitivity, specificity, Matthews correlation coefficient (MCC), Gmean, diagnostic accuracy, Youden's index, positive predictive value, negative predictive value and F1-score criteria. Results: According to CBA model results, sensitivity is 100%, specificity 98%, accuracy 98.6%, balanced accuracy 99%, Youden's index 98%, MCC 96.7%, diagnostic accuracy 98.6%, G-mean 97.7%, negative predictive value 1%, positive predictive value 95.5%, and F1 score 97.7%. According to RCAR model results, sensitivity is 90.5%, specificity 98%, accuracy 95.8%, balanced accuracy 94.3%, Youden's index 88.5%, MCC 89.8%, diagnostic accuracy 95.8%, G-mean 95.6%, negative predictive value 96.2%, positive predictive value 95%, and F1 score 92.7%. Conclusion: When the results are examined, it can be said that the CBA model is more successful in classifying cervical cancer compared to the RCAR model. In addition, the relational classification models created in this study and the rules obtained regarding the disease are promising in terms of their use in early diagnosis and preventive medicine practices for cervical cancer.