Prediction and Detection of Cervical Malignancy Using Machine Learning Models.

Area of Science:

• Oncology
• Medical Informatics
• Public Health

Background:

• Cervical cancer is significantly influenced by human papillomavirus and other predictive factors.
• Early prediction and diagnosis are crucial for preventing cervical cancer.
• Analyzing cervical cancer predictors and managing unbalanced datasets are key challenges in ML applications.

Purpose of the Study:

• To identify and analyze predictors of cervical cancer.
• To evaluate the performance of various machine learning (ML) algorithms in handling unbalanced datasets for cervical cancer prediction.

Main Methods:

• A multi-stage sampling strategy recruited 501 participants for a study involving video-assisted counseling and Pap smear screening.
• Machine learning classification methods including Decision Tree (DT), Random Forest (RF), Logistic Regression (LR), Multi-layer Perceptron (MLP), and Naive Bayes (NB) were employed.
• Models were used to evaluate unbalanced input and target datasets for cervical cancer prediction.

Main Results:

• Out of 501 women, 298 underwent cervical screening, with 26 showing abnormal Pap tests leading to biopsy.
• Seven women were diagnosed with cervical cancer.
• Logistic Regression (LR) demonstrated 88%-94% sensitivity and 84%-89% accuracy, while Decision Tree (DT) showed 83%-84% sensitivity and 84%-88% accuracy in prediction models.

Conclusions:

• Logistic Regression and Decision Tree algorithms were identified as the best-performing ML classifiers for detecting significant cervical cancer predictors.
• These ML models show promise in improving early detection and prediction of cervical cancer, even with imbalanced data.