The performance of a machine learning model on new, unseen data can deviate from its performance on the training data. These deviations, also known as generalization errors, arise due to several factors in the model and the data. In the context of AutoML Vision, a powerful tool provided by Google Cloud for image classification tasks,