Comparative analysis of decision tree-based machine learning algorithms: a trade-off between accuracy and interpretability

Abstract

This paper presents a comparative analysis of decision tree (DT)-based machine learning (ML) algorithms using 11 publicly available datasets from the knowledge extraction based on evolutionary learning (KEEL) repository. The study focuses on two critical aspects of DT-based algorithms: interpretability and accuracy. These measures often conflict, making it challenging to optimize both simultaneously. The research examines the trade-offs between accuracy and interpretability in three algorithms: Java implementation of the repeated incremental pruning to produce error reduction algorithm (JRip), projective and recurring trees (PART), and decision table. Experimental results demonstrate that JRip achieves the highest accuracy among the three algorithms across the datasets. However, decision table exhibits the highest interpretability, followed by PART, while JRip scores the lowest due to its pruning mechanism. These findings provide valuable insights into the balance between accuracy and interpretability, assisting researchers and practitioners in selecting the most suitable algorithm for classification tasks. The findings show a definite trade-off between accuracy and interpretability. Although JRip generates more complicated rule sets, it consistently delivers better accuracy across the majority of datasets. Decision tables offer simpler and easier-to-understand models, despite their lower accuracy.