During an image classification training job, an engineer observes class imbalance in the dataset. Which action should the engineer take to address the imbalance?

Random oversampling is an effective technique to address class imbalance by duplicating samples from the minority class. This increases the representation of the underrepresented class, helping the model learn its characteristics better and preventing it from being biased towards the majority class. Reducing the dataset size is generally counterproductive as it removes valuable information. Transforming some images can augment the dataset, but without specifically targeting the minority class, it might not fully resolve the imbalance. Random data splitting is a standard practice for creating training and validation sets but does not inherently address class imbalance within those sets.