AUC-ROC score is more robust than Accuracy for imbalanced classes.

Sure, I can explain why AUC-ROC score is more robust than Accuracy for imbalanced classes.

1. Imbalanced Classes: In many real-world classification problems, we often encounter imbalanced classes. It's when the number of instances of one class is much higher than the other class. For example, in a dataset of 1000 instances, 950 are 'non-spam' and only 50 are 'spam'. This is an imbalanced dataset.

2. Accuracy: Accuracy is the most intuitive performance measure. It is simply a ratio of correctly predicted observation to the total observations. However, accuracy is not a good measure when the classes are imbalanced. In our example, even if we classify all instances as 'non-spam', we will still get a high accuracy of 95%.

3. AUC-ROC: The Receiver Operating Characteristic (ROC) curve is another common tool used with binary classifiers. The dotted line represents the ROC curve of a purely random classifier; a good classifier stays as far away from that line as possible (toward the top-left corner). The Area Under the Curve (AUC) provides an aggregate measure of performance across all possible classification thresholds. It measures the entire two-dimensional area underneath the entire ROC curve.

4. Robustness: AUC-ROC is more robust than accuracy because it considers all thresholds, it's not just about making the cut. It measures the quality of the model's predictions irrespective of what threshold is chosen, and this is especially important when dealing with imbalanced datasets.

So, in conclusion, when dealing with imbalanced classes, AUC-ROC is a more robust measure than accuracy.