Predictive Analysis of Cervical Cancer Using Machine Learning Techniques

Precautionary measures are less expensive than medical therapy in nearly every country. Any disease that is discovered early on has a higher probability of successfully treating its patient than one that is discovered later in its development. Any therapy we could provide them with would be helpful and would make their lives more pleasant if we did not know how to cure them. Cervical cancer is one of these diseases; it is the fourth-most common kind among women globally. The use of hormonal contraceptives and age are two of the numerous variables that raise the risk of cervical cancer. Cervical cancer mortality rates decrease, and recovery rates are increased with early diagnosis. The goal of this study is to develop a model that can sensitively and correctly detect cervical cancer using machine learning techniques. The voting mechanism will be used that integrates three classifiers logistic regression, decision tree, and random forest. The imbalanced dataset issue was resolved by using SMOTE in conjunction with principal component analysis (PCA) to eliminate dimensions that have no bearing on model accuracy. Next, to avoid the overfitting issue, a stratified tenfold cross-validation procedure was employed. The four target variables in this dataset-Hinselmann, Cytology, Schiller, and Biopsy-are linked to 32 risk factors. For each of the four target variables, we discovered that applying the voting classifier, SMOTE, and PCA approaches helped increase the prediction models' accuracy, ROC-AUC, and sensitivity to greater rates. Accuracy, PPA, and sensitivity ratios increased in the SMOTE-voting model for all target variables by 2.45-5.74%, 2.33-26.84%, and 33.98-42.54%, respectively.