Comparison of Bayesian and classical approaches of logistic regression in modeling risk of preterm birth in Nasarawa State of Nigeria

Abstract

The accuracy of a predictive model is crucial in various fields, including classification tasks. One commonly used method for classification is logistic regression, which relies on maximumlikelihood estimation. However, there is growing interest in exploring the use of Bayesian logistic regression as an alternative approach. This interest stems from the advantages offered by the Bayesian network approach, which allows for explicit modeling of feature dependencies and the introduction of hidden nodes. Furthermore, Bayesian inference can be associated with cognitive processes, making it a potentially powerful tool for analyzing complex data. In a comparative analysis, both classical and Bayesian logistic regression models were evaluated for their performance in classification tasks using data collected from a hospital based retrospective study on postpartum mothers and their babies is confined to Two (2) Tertiary Facilities and Three (3) Secondary Facilities across the three (3) Senatorial Zones of Nasarawa state, Nigeria. The Cohort design is adopted for the study. Model prediction Measures such as R-Square, Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC) were used. Model Performance Measures such as Accuracy, Precision, Recall, F1 Score, Area Under Curve (AUC) were also used. Conclusively, The Bayesian Logistic Regression Model outperforms the classical Logistic Regression Model across all evaluated metrics. It demonstrates higher accuracy, precision, recall, F1-Score, and AUC, indicating better overall predictive performance.