Essay

Improving an Underperforming Fruit Classifier

Question: Suppose a fruit-ripeness classifier is not accurate enough. Describe three improvement ideas, drawn from data-related or model-related choices, and explain how each one could help.

Sample answer: A team could improve the classifier in several ways. First, it could collect more labeled fruit images so the model has more examples to learn from. Second, it could expand the training set with more varied cases, such as different lighting, camera angles, backgrounds, and fruit varieties, so the model better matches real-world use. Third, it could adjust the model setup by trying a larger network if the model seems too simple, a smaller network or stronger regularization if it is overfitting, or more training time if optimization has not yet converged.

Key points:

  • Names at least three valid improvement strategies.
  • Explains that more data can reduce generalization error.
  • Explains that more varied data can reduce mismatch between training and deployment conditions.
  • Explains that changing model capacity, regularization, or training time can address underfitting or overfitting.

Rubric: The response should identify three valid improvement ideas, such as collecting more examples, making the data more diverse, changing model size or architecture, training longer, or adding regularization, and explain how each one improves performance.

0

1

Updated 2026-08-12

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Strategy

Machine Learning Yearning @ DeepLearning.AI