Forest-ORE: Mining Optimal Rule Ensemble to interpret Random Forest models

TL;DR

Forest-ORE tackles the interpretability gap of Random Forests by extracting an optimal rule ensemble through a mixed-integer program that explicitly trades off predictive performance, coverage, and rule complexity. It introduces a four-stage pipeline—Rule Extraction, Rule PreSelection, Rule Selection (MIP), and Rule Enrichment via metarules—to produce compact, informative rule sets while maintaining fidelity to the RF. Validated on 36 benchmark datasets with ten-fold Monte Carlo CV, Forest-ORE achieves competitive accuracy and superior interpretability coverage and rule-size trade-offs compared to baselines, while providing rich metrics for visualization and debugging. The framework offers a practical path toward global and local interpretability of RF models, with potential extensions to regression and other tree ensembles, and invites user-driven parameter tuning to balance accuracy against interpretability in real-world deployments.

Abstract

Random Forest (RF) is well-known as an efficient ensemble learning method in terms of predictive performance. It is also considered a Black Box because of its hundreds of deep decision trees. This lack of interpretability can be a real drawback for acceptance of RF models in several real-world applications, especially those affecting one's lives, such as in healthcare, security, and law. In this work, we present Forest-ORE, a method that makes RF interpretable via an optimized rule ensemble (ORE) for local and global interpretation. Unlike other rule-based approaches aiming at interpreting the RF model, this method simultaneously considers several parameters that influence the choice of an interpretable rule ensemble. Existing methods often prioritize predictive performance over interpretability coverage and do not provide information about existing overlaps or interactions between rules. Forest-ORE uses a mixed-integer optimization program to build an ORE that considers the trade-off between predictive performance, interpretability coverage, and model size (size of the rule ensemble, rule lengths, and rule overlaps). In addition to providing an ORE competitive in predictive performance with RF, this method enriches the ORE through other rules that afford complementary information. It also enables monitoring of the rule selection process and delivers various metrics that can be used to generate a graphical representation of the final model. This framework is illustrated through an example, and its robustness is assessed through 36 benchmark datasets. A comparative analysis of well-known methods shows that Forest-ORE provides an excellent trade-off between predictive performance, interpretability coverage, and model size.

Figures (11)

• Figure 1: Forest-ORE framework for interpreting RF
• Figure 2: XOR dataset: On the left: XOR truth table. On the right: the number in each box refers to the number of instances respecting the condition defined by the box and the color refers to the box target class.
• Figure 3: Upset visualization on XOR rule sets
• Figure 4: Boxplots of the mean accuracy on the testing sets over the benchmark datasets. On the left: Global accuracy. On the right: Accuracy on the testing sets covered by the classifier rules.
• Figure 5: Boxplots of the mean macro precision on the testing sets over the benchmark datasets. On the left: Global macro precision. On the right: macro precision on the testing sets covered by the classifier rules.