Abstract:An important factor in the practical implementation of optimization models is the acceptance by the intended users. This is influenced among other factors by the interpretability of the solution process. Decision rules that meet this requirement can be generated using the framework for inherently interpretable optimization models. In practice, there is often uncertainty about the parameters of an optimization problem. An established way to deal with this challenge is the concept of robust optimization. The goal of our work is to combine both concepts: to create decision trees as surrogates for the optimization process that are more robust to perturbations and still inherently interpretable. For this purpose we present suitable models based on different variants to model uncertainty, and solution methods. Furthermore, the applicability of heuristic methods to perform this task is evaluated. Both approaches are compared with the existing framework for inherently interpretable optimization models.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is: in optimization models, how to create surrogate models that are both robust and easy to interpret to deal with the problem of parameter uncertainty. Specifically, the author hopes to develop decision trees that can handle parameter perturbations as surrogate models in the optimization process, while ensuring that these decision trees are inherently interpretable. The following are the core problems and goals of the paper: ### Core Problems of the Paper 1. **Combination of Interpretability and Robustness**: - Traditional optimization models are usually difficult for users to understand and accept, especially when facing parameter uncertainty. - The author hopes to improve users' trust and acceptance of optimization results by constructing robust and interpretable surrogate models. 2. **Parameter Uncertainty**: - In practical applications, the parameters of optimization problems are often uncertain (for example, measurement errors or inaccurate data). - The author proposes using robust optimization methods to deal with this uncertainty, ensuring that surrogate models can still provide high - quality solutions when facing perturbations. 3. **Application of Decision Trees**: - Decision trees, due to their simple structure and easy - to - understand characteristics, are an ideal choice for constructing interpretable surrogate models. - However, traditional decision trees may perform poorly when facing parameter perturbations, so new methods need to be developed to enhance their robustness. ### Specific Goals - **Construct Robust Decision Trees**: By introducing budgeted uncertainty sets, ensure that decision trees can still provide optimal or near - optimal solutions in the case of parameter perturbations. - **Optimize Model Structure**: Research different types of budgeted uncertainty sets (global and local budgets) and develop corresponding exact and heuristic solution methods. - **Evaluate Performance**: Through computational experiments, compare the performance differences between robust decision trees and traditional nominal decision trees under different uncertainty conditions. ### Main Contributions - Proposed a framework for generating robust and interpretable decision trees as surrogate models in the optimization process. - Developed iterative solution methods and multiple heuristic algorithms to effectively handle uncertainty in large - scale optimization problems. - Proven through experiments that robust decision trees can significantly improve the adaptability to perturbed data while ensuring a certain nominal performance. ### Summary This paper aims to solve the interpretability and robustness problems of optimization models when facing parameter uncertainty. By constructing robust and interpretable decision trees, it improves users' trust and acceptance of optimization results. This research not only helps to improve the practical application effects of optimization models but also provides new ideas and methods for further research in related fields.

Towards Robust Interpretable Surrogates for Optimization

Feature-Based Interpretable Surrogates for Optimization

A Framework for Inherently Interpretable Optimization Models

Multistage Robust Mixed-Integer Optimization under Endogenous Uncertainty

A Robust Optimization Model For Project Portfolio Selection With Information Uncertainty

Solving multiobjective optimization problems with decision uncertainty: an interactive approach

A Note on Piecewise Affine Decision Rules for Robust, Stochastic, and Data-Driven Optimization

New Robust Optimization Approach Induced by Flexible Uncertainty Set: Optimization under Continuous Uncertainty

Modulating Surrogates for Bayesian Optimization

Robust Stochastic Optimization Made Easy with RSOME

A Unified Framework for Adjustable Robust Optimization with Endogenous Uncertainty

Leveraging Model-based Trees as Interpretable Surrogate Models for Model Distillation

Surrogate-Based Simulation Optimization

An Interpretable Rule Creation Method for Black-Box Models based on Surrogate Trees -- SRules

Expected Improvement versus Predicted Value in Surrogate-Based Optimization

Piecewise constant decision rules via branch-and-bound based scenario detection for integer adjustable robust optimization

When stakes are high: balancing accuracy and transparency with Model-Agnostic Interpretable Data-driven suRRogates

Pragmatic Distributionally Robust Optimization for Simple Integer Recourse Models

Optimization under Decision-Dependent Uncertainty

A survey of decision making and optimization under uncertainty

Advances in surrogate based modeling, feasibility analysis, and optimization: A review