Cervical cancer perceived behavioral risk factors using logistic regression technique.

Area of Science:

• Oncology
• Biostatistics
• Machine Learning in Healthcare

Background:

• Cervical cancer poses a significant global health burden, particularly in middle-income nations due to screening challenges.
• Ineffective screening programs contribute to the high incidence of cervical cancer in underserved populations.
• Behavioral risk factors play a crucial role in cervical cancer development and progression.

Purpose of the Study:

• To forecast cervical cancer incidence by analyzing behavioral risk factors.
• To develop and evaluate a robust predictive model for early cervical cancer detection.
• To compare the performance of logistic regression against other machine learning algorithms for cervical cancer prediction.

Main Methods:

• Logistic regression model applied to analyze behavioral risk factors.
• Principal Component Analysis (PCA) used for feature engineering, reducing dimensionality.
• Stratified K-fold cross-validation implemented for balanced class representation.
• L1 regularization incorporated to enhance logistic regression model performance.

Main Results:

• Principal Component Analysis (PCA) condensed data into ten components, explaining 89% of variance.
• The L1-regularized logistic regression model achieved high performance metrics: 97.2% accuracy, 98.1% AUC, 97.2% F1 score, 96.1% specificity, and 0.17 log loss.
• Logistic regression demonstrated superior accuracy (97.2%) compared to Decision Trees, Random Forest, XGBoost (all 93.33%), Naive Bayes (91.67%), and non-regularized logistic regression (87.55%).

Conclusions:

• Early prediction of cervical cancer using behavioral risk factors is crucial for improving outcomes.
• The study presents a reliable and implementable workflow for enhancing cervical cancer classification accuracy.
• Future research should focus on refining predictive models to address social and behavioral barriers in prevention, especially in vulnerable groups.