Evaluating Tree-Based and Probability-Based Learning in Liver Disease Prediction: A Comparison of Decision Tree and Naïve Bayes Algorithms

Abstract

INTRODUCTION: Liver disease imposes a heavy global burden, particularly in developing countries such as Nigeria, where early detection through traditional methods remains challenging due to resource constraints and late symptom onset. This study addresses this by evaluating tree-based versus probability-based algorithms for liver disease prediction.

OBJECTIVES: The specific objectives are to: explore and preprocess liver disease patient dataset from Kaggle using label encoding, and standard scaling techniques; implement and evaluate standalone Decision Tree and Naïve Bayes classifiers via metrics including accuracy, precision, recall, F1-score; perform comparative analysis between the two models; and validate the model's generalizability and propose deployment guidelines for healthcare systems.

METHODS: Techniques such as label encoder, standard scaler were employed for categorical variable conversion, and data scaling, respectively. Tree-based (Decision Tree) algorithm, and Probability-based (Naïve Bayes) algorithm were employed to develop predictive models for liver disease prediction, evaluated using accuracy, precision, recall, and F1-score metrics,

RESULTS: Results reveal that Decision Tree achieved 98% across metrics, outperforming Naïve Bayes (66% accuracy, 74% precision, 66% recall, and 63% f1-score).

CONCLUSION: These results affirm tree-based superiority yet highlight hybrids' potential for efficient, and effective diagnostics. Therefore, this study suggests future work to explore advanced hybrids such as stacking Decision Tree-Naïve Bayes, deep feature engineering, and validation on diverse datasets for clinical deployment in high-burden regions.