MLA-C01 Question 177
Select 3A data science team has deployed a machine learning model for predicting product demand using Amazon SageMaker. However, the model's performance on new data is not meeting business requirements. Which of the following methods could help improve the model's performance?
- A
Increase the size of the training dataset by adding more labeled examples.
- B
Tune hyperparameters such as learning rate and batch size using SageMaker Automatic Model Tuning.
- C
Switch from the current algorithm to a completely different one, such as from linear regression to a neural network, without further evaluation.
- D
Perform feature engineering, such as normalizing numerical features or encoding categorical features.
- E
Reduce the size of the training dataset to focus on higher-quality examples.
Show answer and explanation
Correct answers: A, B, D
Explanation
Improving a machine learning model's performance typically involves increasing the quality or quantity of training data, optimizing the model's hyperparameters, and engineering better input features. These methods directly address common issues such as underfitting, overfitting, or poor generalization. However, switching algorithms or reducing the training dataset without proper evaluation can introduce additional problems and is not recommended in this scenario.
- A. Correct.
Increasing the size of the training dataset can help the model generalize better, especially if the current dataset is small or not representative of the problem domain.
- B. Correct.
Hyperparameter tuning can significantly improve model performance by finding the optimal set of hyperparameters for training.
- C. Incorrect.
Switching to a completely different algorithm without evaluation is not a recommended approach. Algorithm selection should be based on experimentation and analysis of the problem.
- D. Correct.
Feature engineering can enhance model performance by providing the algorithm with better input representations of the data.
- E. Incorrect.
Reducing the size of the training dataset may lead to underfitting and poorer generalization unless the dataset contains noisy or irrelevant data, which is not mentioned in the scenario.