Year: 2026
Type: preprint
Abstract: Background: Classification problems are fundamental in statistical learning and arise in numerous fields such as healthcare, finance, and environmental sciences. Traditional statistical models provide interpretable parameter estimates but may struggle to capture complex nonlinear relationships, whereas machine learning methods often achieve strong predictive performance at the cost of interpretability. These limitations motivate the development of hybrid approaches that integrate probabilistic modeling with modern machine learning techniques. Methods: This study proposes a Bayesian–Machine Learning Ensemble (BMLE) framework that combines Bayesian logistic regression with a machine learning component to improve classification performance while maintaining interpretability. A simulation study was conducted to evaluate the proposed model and compare its performance with a conventional logistic regression model. Model performance was assessed using several classification metrics, including accuracy, precision, recall, F1-score, and the area under the receiver operating characteristic curve (AUC). In addition, parameter estimation was evaluated using bias, root mean squared error (RMSE), and coverage probability. Results: The simulation results demonstrate that the proposed BMLE model consistently outperforms the conventional logistic regression model across all predictive performance metrics. The corrected model achieved higher accuracy, precision, recall, and F1-score, as well as a higher AUC value, indicating improved discriminative ability. Furthermore, the Bayesian component provided stable parameter estimates and meaningful uncertainty quantification through credible intervals. Conclusion: The proposed Bayesian–Machine Learning Ensemble framework offers a flexible and robust approach for classification problems by integrating probabilistic inference with machine learning techniques. The results suggest that the hybrid modeling strategy improves predictive accuracy while preserving interpretability, making it a promising methodology for complex classification tasks.
Source: Research Square
Author: Romuald Daniel BOY-NGBOGBELE
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7160495672