Databricks Machine Learning Associate Question 616
Select 3You are training a machine learning model using Databricks and observe that the training accuracy is very high, but the test accuracy is significantly lower. Which of the following actions would most likely improve the model's performance on the test dataset while addressing the bias-variance tradeoff?
- A
Decrease the model's complexity by reducing the number of features or using regularization.
- B
Increase the model's complexity by adding more layers or parameters.
- C
Gather more labeled training data to better represent the underlying distribution.
- D
Reduce the amount of training data to focus on high-quality samples.
- E
Use cross-validation to ensure that the model generalizes well to unseen data.
Show answer and explanation
Correct answers: A, C, E
Explanation
The scenario described indicates a high variance problem, where the model overfits the training data and performs poorly on unseen test data. Reducing model complexity, gathering more training data, and using cross-validation are effective strategies to address overfitting and improve the model's generalization performance. These actions collectively balance the bias-variance tradeoff.
- A. Correct.
Decreasing the model's complexity can reduce overfitting, which is likely the cause of the high training accuracy but low test accuracy. This addresses the variance in the bias-variance tradeoff.
- B. Incorrect.
Increasing the model's complexity would likely exacerbate the overfitting problem, as the model may memorize the training data even more.
- C. Correct.
Gathering more labeled training data can help the model better learn the underlying patterns and reduce overfitting, thus addressing the variance issue.
- D. Incorrect.
Reducing the amount of training data is counterproductive, as it limits the model's ability to learn the underlying patterns and could increase both variance and bias.
- E. Correct.
Using cross-validation ensures that the model's performance is evaluated on different subsets of data, helping to improve generalization and address overfitting.