nerdexam
Amazon

MLS-C01 · Question #298

A machine learning (ML) specialist is training a multilayer perceptron (MLP) on a dataset with multiple classes. The target class of interest is unique compared to the other classes in the dataset…

The correct answer is A. Add class weights to the MLP's loss function, and then retrain. The ML specialist needs to improve recall for a unique, underperforming target class in an MLP model without significant time investment.

Modeling

Question

A machine learning (ML) specialist is training a multilayer perceptron (MLP) on a dataset with multiple classes. The target class of interest is unique compared to the other classes in the dataset, but it does not achieve an acceptable recall metric. The ML specialist varies the number and size of the MLP's hidden layers, but the results do not improve significantly. Which solution will improve recall in the LEAST amount of time?

Options

  • AAdd class weights to the MLP's loss function, and then retrain.
  • BGather more data by using Amazon Mechanical Turk, and then retrain.
  • CTrain a k-means algorithm instead of an MLP.
  • DTrain an anomaly detection model instead of an MLP.

How the community answered

(46 responses)
  • A
    76% (35)
  • B
    9% (4)
  • C
    13% (6)
  • D
    2% (1)

Why each option

The ML specialist needs to improve recall for a unique, underperforming target class in an MLP model without significant time investment.

AAdd class weights to the MLP's loss function, and then retrain.Correct

Adding class weights to the loss function directly addresses class imbalance, a common cause of low recall for minority classes. By assigning a higher weight to the target class, the model is penalized more for misclassifying it, leading to improved recall for that specific class without requiring extensive data collection or model architecture changes, making it a quick solution.

BGather more data by using Amazon Mechanical Turk, and then retrain.

Gathering more data is a valid strategy for model improvement but is typically a time-consuming process, which contradicts the requirement for the LEAST amount of time.

CTrain a k-means algorithm instead of an MLP.

Training a k-means algorithm, an unsupervised clustering method, is not suitable for improving recall in a supervised multi-class classification problem.

DTrain an anomaly detection model instead of an MLP.

Training an anomaly detection model is designed for identifying rare outliers, not for enhancing the recall of a specific target class within a standard classification task.

Concept tested: Class imbalance handling in machine learning

Source: https://learn.microsoft.com/en-us/azure/machine-learning/concept-automl-imbalance

Topics

#Class Imbalance#Recall Metric#Multilayer Perceptron#Model Optimization

Community Discussion

No community discussion yet for this question.

Sign up to join the discussion
Full MLS-C01 Practice