Evaluating the Predictive Performance and Computational Efficiency of Logistic Regression and Naïve Bayes for Diabetes Classification

Abstract

Diabetes is a growing global health concern, affecting millions of individuals and placing a significant burden on healthcare systems. Early and accurate diagnosis is crucial for effective disease management, yet traditional diagnostic methods may be time-consuming or less efficient in handling large volumes of patient data. Machine learning algorithms, such as Logistic Regression Model (LRM) and Naive Bayes (NBayes), offer potential solutions for improving the accuracy and speed of diabetes classification based on available datasets. However, there is limited research on comparing the effectiveness of these two widely used algorithms in predicting diabetes dataset, particularly in terms of their performance metrics such as accuracy, precision, recall, and computational efficiency. This study aims to address this gap by evaluating and comparing the performance of LRM and NBayes in predicting the presence or absence of diabetes, utilizing a publicly available dataset on the WEKA platform. The results showed that LRM outperforms NBayes in terms of prediction accuracy, correctly classifying 15,081 instances compared to NBayes' 14,748. LRM also achieves superior performance across multiple metrics, including accuracy (88.71% vs 86.75%), precision, recall, F-measure, and ROC Area, while maintaining a lower error rate (11.29% vs 13.25%). However, NBayes exhibits faster model build time (0.09 sec vs 1.17 sec). Based on these findings, LRM is recommended for diabetes dataset classification. Future research may explore comparisons of LRM with other algorithms such as Support Vector Machines, Decision Trees, and Neural Networks to assess LRM's generalizability and robustness