Optimal Interpretable Decision Trees via Integer Linear Programming
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing prediction models, such as deep neural networks and traditional decision trees, face challenges in interpretability and optimization in process control environments, leading to difficulties in understanding decision-making processes and achieving accurate, scalable solutions.
Innovation Solution
The development of optimal interpretable decision trees using integer linear programming techniques, which involve data subset selection through linear programming and mixed-integer programming formulations to build and optimize decision trees, enhancing interpretability, tractability, and training accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep neural networks are used for prediction modeling, then prediction accuracy is improved, but interpretability deteriorates
Solution Approach 1:
The patent segments the data processing task into two distinct phases: (1) using deep neural networks for accurate predictions, and (2) using separate interpretability models (LIME, SHAP, counterfactuals) to explain specific predictions. This segmentation allows each component to optimize for its specific function without compromise.
Solution Approach 2:
The patent introduces intermediary explanation models that act as mediators between the black-box DNN and the user. These intermediaries (local linear models, attribution methods, counterfactual generators) translate complex neural network decisions into human-understandable explanations without modifying the original accurate prediction model.
2Loss of information
If traditional decision trees are used for interpretability, then interpretability is improved, but prediction accuracy deteriorates
Solution Approach 1:
The patent merges the strengths of multiple approaches by combining deep neural networks (for accuracy) with traditional decision tree interpretability methods (for explanation). The system integrates DNN prediction capabilities with post-hoc explanation techniques that use decision-like reasoning to maintain both accuracy and interpretability.
Solution Approach 2:
The patent performs preliminary data subset selection using linear programming before building the final predictive model. This preliminary action identifies and removes redundant or noisy features, enabling simpler models to achieve higher accuracy while maintaining interpretability.
3Device complexity
If data subset selection is performed to improve model simplicity, then model simplicity is improved, but training data utilization deteriorates
Solution Approach 1:
The patent changes the parameter selection criteria from traditional accuracy-based metrics to linear programming objective functions that explicitly optimize for model simplicity and data subset quality. By changing the optimization parameters, the system identifies minimal data subsets that maintain predictive power while enabling simpler models.
Solution Approach 2:
The patent performs preliminary data subset selection using linear programming to identify the most informative training samples before model training. This preliminary filtering ensures that the reduced dataset captures essential patterns, allowing simpler models to achieve comparable performance without significant data loss.
Data Source
AI summary
Aspects of the invention include an optimal interpretable decision tree using integer linear programming techniques. A non-limiting example computer-implemented method includes receiving, using a processor, a plurality of data inputs from a process and selecting, using the processor, a data subset from the plurality of data inputs by solving linear programming to obtain a solution. The method builds and optimizes, using the processor, an optimal decision tree based on the data subset and alerts, using the processor, a user when a prediction of the optimal decision tree is greater than a threshold value.