Case context: Your team has spent several weeks creating synthetic vibration traces for a factory sensor classifier. After adding the synthetic traces to the training set, the model’s performance on the development set has not improved. A few teammates want to stop the project and move on.

Question: Based on Andrew Ng’s guidance, what is the most likely problem, and what should the team do next?

Sample answer: The most likely problem is that the synthetic traces are still not detailed enough to match the real data distribution. Andrew Ng’s advice is that this kind of work can require weeks of refinement before it starts to help. So the team should not quit just because the first attempts failed; instead, they should inspect what is missing, adjust the generation process, and keep improving the synthetic data so it better resembles the real distribution. If they get that match right, the payoff can be a much larger effective training set.

Key points:

• The synthetic data likely does not yet match the real distribution closely enough.
• It is normal for this process to take weeks before showing improvement.
• The team should refine the synthetic data rather than abandon the effort.

Rubric: The response should identify that the synthetic data likely still differs too much from the real distribution, acknowledge that several weeks of iteration without improvement can be normal, and recommend continuing to refine the data instead of dropping the project.