Cervical Cancer Prediction Using Machine Learning Techniques

Precautionary measures are often less expensive than medical care in practically all nations. Early detection of any disease increases the likelihood that it will be effectively treated compared to late detection. Since no one has any training in medicine, any therapy would be helpful and make life more comfortable. Cervical cancer is one such condition that is the fourth most prevalent malignancy in women globally. Age and hormonal contraception use are two elements that raise the risk of developing cervical cancer. The ability to diagnose cervical cancer increases survival rates and lowers the mortality rate. To employ machine learning, this paper methods to find a cervical cancer diagnostic model with great sensitivity and precision. By combining three classifiers Logistic Regression, Decision Tree, and Random Forest into a voting mechanism, the classification model was developed using the UCI cervical cancer risk factor dataset. The imbalance dataset issue was addressed by combining the principal component analysis (PCA) method with the SMOTE to minimize dimensions that have little bearing on the model's precision. Then, to avoid overfitting, stratified tenfold cross-validation was applied. Along with the four-goal variables Schiller, Hinselmann, Cytology, and Biopsy, this dataset also includes 32 risk factors. With the use of the voting classifier, SMOTE, and PCA techniques, it was discovered that the sensitivity, accuracy, and the prediction models created for each of the four target variables saw an improvement in their ROC AUC. Reduced dimensions that do not affect model accuracy were used in conjunction with the principal component analysis (PCA) strategy and the SMOTE.