Explanation-Aware Automated Machine Learning
This repository accompanies the research paper:
“Multi-Objective Automated Machine Learning for Explainable Artificial Intelligence: Optimizing Predictive Accuracy and Shapley-Based Feature Stability.”
In high-stakes domains such as agriculture, machine learning models must be not only accurate but also transparent and aligned with domain knowledge. This project presents a novel multi-objective optimization framework that jointly maximizes predictive performance and explanation stability. Specifically, we introduce a formal metric based on the variance of Shapley Additive Explanations across cross-validation folds, embedding it directly into the model selection process.
Our approach leverages the Non-dominated Sorting Genetic Algorithm II to evolve models that balance predictive accuracy with robust, semantically consistent explanations. When applied to potato yield prediction, the framework outperforms both H2O.ai's Automatic Machine Learning platform and traditional grid search, producing models that are both high-performing and interpretable.
🔍 Key Features
- Multi-objective optimization for predictive accuracy and explanation stability
- Shapley-based metric embedded into the model selection loop
- Implementation using NSGA-II for evolutionary search
- Reproducible case study in potato yield forecasting
- Baseline comparisons with grid search and H2O.ai’s platform