Cervical Cancer Identification with Synthetic Minority Oversampling Technique and PCA Analysis using Random Forest...

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Cervical cancer is a leading global malignancy in women, often asymptomatic in early stages.
• Risk factors include human papillomavirus (HPV), STDs, and smoking, necessitating effective detection methods.

Purpose of the Study:

• To construct a robust classification model for cervical cancer detection using identified risk factors.
• To evaluate the efficacy of Random Forest (RF) combined with data balancing and feature reduction techniques.

Main Methods:

• Utilized a dataset with 32 risk factors and four diagnostic variables (Hinselmann, Schiller, Cytology, Biopsy).
• Employed Random Forest (RF) classification, Synthetic Minority Oversampling Technique (SMOTE) for data imbalance, and Recursive Feature Elimination (RFE) & Principal Component Analysis (PCA) for feature reduction.
• Developed an RSOnto ontology to visualize classification performance improvements.

Main Results:

• The SMOTE technique effectively addressed data imbalance without compromising diagnostic accuracy.
• Classification metrics (Accuracy, Sensitivity, Specificity, PPA, NPA) remained high across all four diagnostic variables post-SMOTE.
• Feature reduction techniques (RFE, PCA) were integrated to optimize the model's predictive power.

Conclusions:

• The proposed RF model, augmented by SMOTE and feature reduction, demonstrates high accuracy in classifying cervical cancer risk.
• This approach offers a promising tool for early detection and improved patient outcomes in cervical cancer screening.