AWS Certified Machine Learning – Specialty — Question 57

During mini-batch training of a neural network for a classification problem, a Data Scientist notices that training accuracy oscillates.
What is the MOST likely cause of this issue?

Answer options

• A. The class distribution in the dataset is imbalanced.
• B. Dataset shuffling is disabled.
• C. The batch size is too big.
• D. The learning rate is very high.

Correct answer: D

Explanation

A very high learning rate can cause the model to converge erratically, leading to oscillations in training accuracy. An imbalanced class distribution, disabled dataset shuffling, or a large batch size may affect performance but are less likely to cause the specific issue of oscillation in accuracy.