Machine Learning and Bayesian Approaches for Cervical Cancer Risk Prediction: Evidence from Behavioral Risk Factor Analysis with Kenya Contextualization
,
Click an author's name to view their details
Abstract
Background: Cervical cancer remains a significant public health burden globally, with sub-Saharan Africa experiencing disproportionately high mortality rates. Kenya faces particular challenges with limited screening access and high disease incidence. This study applies machine learning and Bayesian statistical methods to identify behavioral and psychological risk factors associated with cervical cancer status. Methods: We analyzed the UCI Cervical Cancer Behavior Risk dataset (n=72) containing 19 behavioral and psychological features. Unsupervised clustering (K-means and hierarchical) was performed to identify behavioral phenotypes. Four classification algorithms (Logistic Regression, Random Forest, Support Vector Machine, and XGBoost) were evaluated using 5-fold stratified cross-validation. Bayesian logistic regression was implemented using PyMC with No-U-Turn Sampler (NUTS) to quantify uncertainty in risk factor effects. Results: Logistic Regression achieved the highest discriminative performance (AUC-ROC = 0.992), followed by SVM (0.981). Clustering analysis identified two distinct behavioral phenotypes, with one cluster showing 60.7% cancer prevalence versus 9.1% in the other. Feature importance analysis consistently identified empowerment-related constructs (abilities, desires, knowledge) and perceived severity as top predictors. Bayesian analysis revealed that empowerment_desires (β = -1.545, 95% HDI: [-2.790, -0.324]) and perception_severity (β = -1.528, 95% HDI: [-2.862, -0.200]) were significantly associated with reduced cancer risk, suggesting protective behavioral patterns. Conclusions: Machine learning models demonstrate excellent discriminative ability for cervical cancer risk stratification based on behavioral factors. The identification of empowerment and perceived severity as protective factors provides actionable targets for intervention. These findings have important implications for risk-based screening prioritization in resource-constrained settings like Kenya, where mobile health applications could deploy these models to enhance early detection efforts.
How to Cite
OWILI, C. A., & Rangita, A. (2026). Machine Learning and Bayesian Approaches for Cervical Cancer Risk Prediction: Evidence from Behavioral Risk Factor Analysis with Kenya Contextualization. Journal of Contemporary Academic Research and Methodologies, 1(7). https://doi.org/10.5281/zenodo.22744109
OWILI, CHRISTINE AKINYI, and Apaka Rangita. "Machine Learning and Bayesian Approaches for Cervical Cancer Risk Prediction: Evidence from Behavioral Risk Factor Analysis with Kenya Contextualization." Journal of Contemporary Academic Research and Methodologies, vol. 1, no. 7, 2026. DOI: https://doi.org/10.5281/zenodo.22744109
OWILI, CHRISTINE AKINYI, and Apaka Rangita. "Machine Learning and Bayesian Approaches for Cervical Cancer Risk Prediction: Evidence from Behavioral Risk Factor Analysis with Kenya Contextualization." Journal of Contemporary Academic Research and Methodologies 1, no. 7 (2026). https://doi.org/10.5281/zenodo.22744109
References
This article is published under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. Free to read, share, and adapt with attribution.