Beyond Explaining Predictions: Logic-Based Explanations for Confidence in Machine Learning Models
Researchers have developed a new method for generating logic-based explanations for machine learning model confidence. This approach, called confidence-aware abductive explanations, ensures that explanations not only preserve the predicted class but also meet a specified confidence threshold. Experiments on boosted trees demonstrated that these new explanations improve minimum guaranteed confidence with only a slight increase in length, making them suitable for applications requiring trustworthy decision-making. AI
IMPACT Enhances trustworthiness in ML applications by providing clearer confidence guarantees for model predictions.