Which methods can help resolve data imbalance problems in deep learning tasks?

Random oversampling is a method where instances from the minority class are duplicated in the training dataset, thereby balancing the class distribution. This approach increases the representation of the minority class without losing any information, which can improve the model's ability to learn patterns associated with this class.

Synthetic sampling refers to a technique where new, synthetic instances of the minority class are created based on existing instances. This is often done using methods like SMOTE (Synthetic Minority Over-sampling Technique), which can create more diverse training samples rather than simply duplicating existing instances. As a result, synthetic sampling helps enhance the learning capacity of the model by providing it with more varied examples of the minority class, ultimately leading to better generalization.

Random undersampling involves reducing the number of instances from the majority class to balance the dataset. While this can effectively rectify imbalance, it risks losing potentially valuable information from the majority class, which could harm the model’s performance.

Batch deletion is not a common method for addressing data imbalance. It does not specifically focus on balancing classes and may have negative implications for data integrity and model training.

In this context, synthetic sampling stands out because it not only addresses the imbalance but also enriches the dataset, making it a robust choice for improving model performance in