MLS-C01 Question 122
Select 3A data scientist is building a machine learning model to predict customer churn for an e-commerce platform. They have chosen an XGBoost algorithm for training. After training the model, they notice that the model's performance on the test dataset is significantly worse than its performance on the training dataset. Which of the following actions should the data scientist take to address this issue?
- A
Use cross-validation to evaluate the model's performance more robustly.
- B
Increase the number of estimators (trees) in the XGBoost model.
- C
Implement regularization techniques like L1 or L2 to reduce model overfitting.
- D
Reduce the maximum depth of the trees in the XGBoost model.
- E
Perform feature scaling on the input dataset to improve model generalization.
Show answer and explanation
Correct answers: A, C, D
Explanation
The significant performance gap between the training and test datasets suggests the model is overfitting. To address overfitting in XGBoost, the data scientist can implement regularization techniques, reduce the maximum depth of trees, and validate the model's performance using cross-validation to ensure robustness. Increasing the number of estimators or applying feature scaling would not mitigate overfitting in this scenario.
- A. Correct.
Using cross-validation can help verify if the model's poor performance is due to overfitting or an issue with the way the test set is split. This is a valid step to take.
- B. Incorrect.
Increasing the number of estimators would likely exacerbate overfitting, as it increases model complexity. This is not a recommended approach in this scenario.
- C. Correct.
Regularization techniques like L1 (Lasso) or L2 (Ridge) can help reduce overfitting by penalizing overly complex models. This is a valid step to improve generalization.
- D. Correct.
Reducing the maximum depth of the trees can prevent the model from overfitting by limiting its complexity. This is a valid approach.
- E. Incorrect.
Feature scaling is not required for tree-based models like XGBoost, as they are scale-invariant. This would not address the overfitting problem.