EXPERIMENTAL COMPARATIVE ANALYSIS OF NAIVE BAYES AND RULE BASED DECISION TREE PERFORMANCE IN ATTRIBUTIVE CLASSIFICATION OF SCHOLARLY ARTICLES

Abstract

Document classification categorizes documents based on their attributes or contents to determine how they will be stored, retrieved, and shared. Improper classification of documents can lead to significant waste of time, resources, productivity, and efficiency in organizational information management systems. This study presents the Experimental Comparative Analysis (ECA) of Naive Bayes Model (NBM) and Rule-Based Decision Tree Model (RBDTM) in the classification of journal articles based on their attributes such as Name of Journal (NOJ) regarded as the specific domain, Peer Reviewed (PR) status, Volume Information (VI), the Title (T), Author Names (AN), Digital Object Identifier (DOI) and Keywords (KW). The models were trained based on direct algebraic solutions, saved parameters, and deterministic passes without gradient descent updates, using supervised Machine Learning (ML) approach and K-fold validation sampling technique for training and testing. Performance metrics such as accuracy, precision, recall, specificity, F1-score, program execution time, information retrieval response time, memory usage, and confusion matrix were used for analysis and evaluation. The training and testing datasets were obtained from open-licensed, peer-reviewed journal articles from Africa, Asia, the United States of America, Eastern and Western Europe. The average test results at the baseline threshold showed that the NBM performed better than the RBDTM in accuracy, recall, and F1-score (68%, 68%, and 80.6%, respectively); both models achieved 100% precision and 0% specificity. In contrast, RBDTM outperformed NBM in program execution time, information retrieval response time, and memory usage (17.19ms, 7.44ms, and 0.13%, respectively). At the optimized threshold, both models tied with 100% accuracy, 100% precision, 100% recall, 0% specificity, and 100% F1-score, but the RBDTM demonstrated the best overall performance with 18.2ms program execution time, 7.74ms information retrieval response time, and 0.14% memory usage.