Comparison of Decision Tree and Gaussian Naive Bayes Algorithms for Classification of Inpatients Based on Hematology Data

Abstract

Background: Hematology data analysis has an important role in supporting clinical decision-making, particularly in determining patient treatment status. However, manual analysis of laboratory data requires considerable time and may increase the risk of errors. Machine learning approaches can assist in identifying clinical patterns from hematology data. Objective: This study aims to compare the performance of Decision Tree and Gaussian Naive Bayes algorithms in classifying inpatient and outpatient status based on hematology laboratory data. Methods: This quantitative experimental study applied a Knowledge Discovery in Databases approach. The dataset consisted of 4,412 hematology examination records obtained from a public dataset. Data preprocessing included data cleansing, feature encoding, and data splitting into 80% training data and 20% testing data. Decision Tree and Gaussian Naive Bayes algorithms were implemented using Python and evaluated using accuracy, precision, recall, and F1-score metrics. Results: The Decision Tree algorithm achieved better performance than Gaussian Naive Bayes, with an accuracy of 73.61%, precision of 66.67%, recall of 69.47%, and F1-score of 68.04%. Meanwhile, Gaussian Naive Bayes obtained an accuracy of 68.63%, precision of 64.71%, recall of 49.30%, and F1-score of 55.96%. Conclusion: Decision Tree demonstrated superior performance in classifying patient treatment status based on hematology data. Its ability to generate interpretable clinical decision patterns makes it more suitable for supporting clinical decision support systems compared with Gaussian Naive Bayes.