Performance evaluation of ovarian cancer detection using on machine learning approaches based on feature selection

Abstract

Ovarian cancer is one of the most dangerous genecology cancers because it does not show any symptoms in the early stages and there aren’t any good tests to find it. Early detection is important for increasing patient survival, but traditional diagnostic methods often don't have the right level of sensitivity and specificity. This study looks into how machine learning (ML) and deep learning (DL) can be used to find ovarian cancer early and accurately. We looked at five models: Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), and Elman Recurrent Neural Network (ERNN). We did this before and after using the RF algorithm to select features.  The results show that feature selection made all of the models work much better. The ERNN model performed the best overall, with accuracy going from 89.8% to 92.5% and AUC-ROC going from 0.94 to 0.96 after feature selection. In the same way, ANN and RF got 92.1% and 91.0% accuracy, respectively, with big improvements in precision, recall, and F1-score. These results show how important it is to optimize features to make models work better. They also confirm that intelligent ML-based systems could be used to reliably find ovarian cancer early.