Ensemble learning, particularly tree-based ensemble techniques, is acknowledged as the advanced approach to solving a wide range of challenging issues because of its exceptional performance in multiple machine learning applications. Nonetheless, even with its capacity for prediction, ensemble learning’s interpretability continues to be a problem, which restricts its use in situations where decision transparency is necessary. We suggest a novel model named TERM (Tree Ensemble Models for Interpretable Rule Mining) in order to get over this restriction and enhance the interpretability of Ensemble Tree Models. This method converts the complex binary classification decision-making process inside tree ensemble models into a series of rule lists that are globally equivalent to the decision-making process of the original model. This transformation aids in a more precise understanding and interpretation of the model’s predictive outcomes, enhancing its interpretability. Compared with existing rule extraction methods, TERM can simultaneously be applied to bagging methods and boosting methods. During the rule extraction phase, an adaptive set cover algorithm ADAPT-TA, is proposed to select rule subsets from the original rule set while avoiding local optima. Additionally, this method reduces the generalization error in the set of rules and runs faster on ensemble models involving multiple decision trees. Through comparisons with six other models, we evaluated the performance and applicability of TERM, demonstrating its outstanding accuracy and interpretability.