Researchers have found a way to interpret the predictions made by gradient-boosted ensembles, which are often used in machine learning. By viewing each leaf value as a coordinate in a high-dimensional space, they've been able to create exact contrastive explanations for why certain instances were predicted one way or another. This approach makes it possible to identify specific features that contributed to the decision, and even provides a method for auditing the model's arithmetic. The researchers tested this new method on several datasets and found it to be effective in providing actionable recommendations. --- Why it matters: This matters because it allows for more transparent and interpretable machine learning models, which is essential for building trust in AI systems. By being able to identify specific features that contribute to a decision, developers can improve their models and make them more reliable. Source: https://arxiv.org/abs/2608.19127