Ensemble Machine Learning Algorithms for the Diagnosis of Cervical Cancer
Fahima Khanam, Md. Rubaiyat Hossain Mondal · 2021
Cervical cancer is a deadly form of cancer and early screening is important for preventing it. In many countries there are limited facilities for the screening of cervical cancer. The paper is centered on the application of machine learning algorithms for the diagnosis of cervical cancer. For this, a dataset is accumulated from the repository of University of California Irvine (UCI).The UCI dataset has 36 attributes or features of 858 samples. The dataset has imbalance in the number of patients and normal samples and hence Borderline Synthetic Minority Over-sampling Technique (Borderline-SMOTE) is used. A single target variable representing whether the individual has cancer or not is formed by using the logical OR operation among the four target attributes namely Hinselmann, Schiller, Cytology, and Biopsy. Next, extra tree classifier method is applied to the dataset to find the most important attributing to cervical cancer. Using the feature importance values, the 10 most important features are determined for classification. A number of base classifiers and several ensemble methods including bagging, boosting and stacking are applied. Results show that stacking with a mixture of random forest, support vector machine, ExtraTreeClassifier, XGBoost, and Bagging has the highest classification accuracy of 94.4%.