NCA-GENL Question 31
Single answerYou are training a machine learning model to classify images of vehicles into categories such as cars, trucks, and motorcycles. After performing feature extraction using color histograms and edge detection, you want to evaluate multiple models to determine the best one for deployment. Which approach would ensure you select the model with the best performance while minimizing overfitting?
- A
Use the model with the highest accuracy on the training dataset.
- B
Split the dataset into training, validation, and test sets, and select the model with the best performance on the validation set.
- C
Train every model on the entire dataset and choose the one that performs best on unseen test data.
- D
Perform cross-validation to evaluate each model’s performance and select the one with the lowest average validation error.
Show answer and explanation
Correct answer: D
Explanation
Cross-validation provides a reliable way to evaluate model performance across multiple subsets of the data. It reduces the risk of selecting a model that performs well on only a specific split of the dataset, ensuring the model generalizes well to unseen data. This approach is essential for selecting the best model while minimizing overfitting.
- A. Incorrect.
Selecting the model based solely on training accuracy does not account for overfitting, as the model may not generalize well to unseen data.
- B. Incorrect.
Using a validation set is a valid approach, but it is less robust than cross-validation when evaluating models, as it relies on a single split, which may not represent the entire data distribution.
- C. Incorrect.
Training on the entire dataset and testing on unseen data is not a standard practice, as it leaves no data for validation during model development and risks overfitting.
- D. Correct.
Cross-validation is a robust method to evaluate models by splitting the dataset into multiple folds, ensuring that performance is measured on various data splits, reducing the risk of overfitting.