NCA-AIIO Question 90
Select 4A data scientist is using an NVIDIA-powered AI infrastructure to train a deep learning model for predicting customer churn. They notice that the training accuracy is high, but the model performs poorly on the test set. Upon further investigation, they suspect a relationship between an outdated dataset and the poor performance. Which actions should they take to identify relationships or trends that may be affecting the model's results?
- A
Conduct exploratory data analysis (EDA) to identify biases or anomalies in the dataset.
- B
Use NVIDIA TensorBoard to visualize training metrics over time.
- C
Ensure the dataset is updated regularly and reflects recent trends.
- D
Fine-tune the model's hyperparameters to improve its generalization.
- E
Analyze feature importance to determine which input variables significantly influence predictions.
Show answer and explanation
Correct answers: A, B, C, E
Explanation
To address poor model performance related to outdated datasets, it is crucial to identify relationships or trends affecting the model. This can be achieved by conducting EDA, visualizing training metrics, ensuring the dataset is current, and analyzing feature importance. These steps provide insights into data quality, feature relevance, and potential biases, which are critical for improving model performance. While fine-tuning hyperparameters is useful, it does not directly tackle the root cause of the issue in this scenario.
- A. Correct.
Conducting exploratory data analysis (EDA) can help uncover biases, anomalies, or outdated trends in the dataset that may be affecting the model's results. This is a critical step in identifying factors impacting performance.
- B. Correct.
Using NVIDIA TensorBoard allows the user to monitor training metrics and detect issues such as overfitting or unstable training, which may help diagnose the problem.
- C. Correct.
Ensuring the dataset is up-to-date is essential for maintaining a model that reflects current trends and avoids making predictions based on outdated information.
- D. Incorrect.
Fine-tuning hyperparameters is a valid approach for improving model performance, but it does not directly address the issue of outdated data or identifying relationships that impact results.
- E. Correct.
Analyzing feature importance helps determine which variables contribute most to the model's predictions, allowing the user to assess whether outdated or irrelevant features are affecting the model's performance.