MLS-C01 Question 82
Single answerYou are developing a machine learning model to classify customer support tickets into predefined categories. During the dataset preparation phase, you find that your dataset contains 10,000 labeled examples spread across 5 categories. However, some categories have only 200 labeled examples while others have over 3,000. What is the best way to determine if you have sufficient labeled data to train a high-performing model?
- A
Perform exploratory data analysis (EDA) and calculate the class imbalance ratio.
- B
Use a small subset of your labeled data to train a baseline model and evaluate its performance.
- C
Increase the number of labeled examples for the underrepresented categories until all categories have approximately equal counts.
- D
Use data augmentation techniques to synthetically increase the size of the dataset.
Show answer and explanation
Correct answer: B
Explanation
To determine if you have sufficient labeled data to train a high-performing model, it is essential to evaluate the model's performance using the data you currently have. Training a baseline model allows you to analyze whether the dataset size and label distribution are adequate or if additional data collection or preprocessing is needed. Other options, such as calculating class imbalance or using data augmentation, are helpful techniques for improving dataset quality but do not directly assess sufficiency.
- A. Incorrect.
While calculating the class imbalance ratio is useful for understanding the distribution of your data, it does not directly help determine if you have sufficient labeled data to train a high-performing model.
- B. Correct.
Training a baseline model on a subset of labeled data allows you to assess whether the current dataset size and quality are sufficient, based on the model's performance metrics. This is the best approach to determine sufficiency.
- C. Incorrect.
Balancing the dataset by collecting more labeled examples is a potential solution for class imbalance but does not necessarily help determine if the total labeled data is sufficient for model training.
- D. Incorrect.
Data augmentation is useful when you have limited data, but it does not directly address the question of whether your current labeled dataset is sufficient to train a high-performing model.