nerdexam
Amazon

MLS-C01 · Question #155

A Data Scientist is working on a predictive maintenance mode and received a company's dataset with 500,000 measurements of machine behavior during both normal operations and failures. In the dataset…

The correct answer is A. Request more data from the company, focusing on failure samples. B. Use an approach to create synthetic samples, such as oversampling. To address a highly imbalanced dataset (98% normal, 2% failures) while minimizing information loss, the data scientist should acquire more samples of the minority class and generate synthetic samples using oversampling techniques.

Modeling

Question

A Data Scientist is working on a predictive maintenance mode and received a company's dataset with 500,000 measurements of machine behavior during both normal operations and failures. In the dataset 98% of the samples were collected during normal operations and 2% were collected during failures. Which of the following actions should address the imbalance while minimizing information loss? (Select TWO.)

Options

  • ARequest more data from the company, focusing on failure samples.
  • BUse an approach to create synthetic samples, such as oversampling.
  • CRemove normal operations samples unti the sample amount matches the number of failure
  • DRun a Latent Dirichlet Allocation (LDA) algorithm on the dataset.
  • ERemove all failure samples and perform classification training using the normal operations

How the community answered

(20 responses)
  • A
    70% (14)
  • C
    10% (2)
  • D
    15% (3)
  • E
    5% (1)

Why each option

To address a highly imbalanced dataset (98% normal, 2% failures) while minimizing information loss, the data scientist should acquire more samples of the minority class and generate synthetic samples using oversampling techniques.

ARequest more data from the company, focusing on failure samples.Correct

Requesting more data for the minority class (failure samples) directly increases its representation, helping to balance the dataset and provide the model with more examples of the critical event without discarding any existing information.

BUse an approach to create synthetic samples, such as oversampling.Correct

Oversampling techniques, such as Synthetic Minority Over-sampling Technique (SMOTE), create synthetic data points for the minority class based on existing samples, effectively balancing the dataset and improving model learning for the underrepresented class while preserving original information.

CRemove normal operations samples unti the sample amount matches the number of failure

Removing normal operations samples (downsampling) to match the minority class size would lead to a significant loss of valuable information from the majority class, potentially making the model generalize poorly.

DRun a Latent Dirichlet Allocation (LDA) algorithm on the dataset.

Latent Dirichlet Allocation (LDA) is a topic modeling algorithm used for discovering abstract topics in a collection of documents, not for addressing class imbalance in a classification dataset.

ERemove all failure samples and perform classification training using the normal operations

Removing all failure samples would eliminate the target class entirely, making it impossible to train a predictive maintenance model to detect failures.

Concept tested: Handling imbalanced datasets

Source: https://developers.google.com/machine-learning/crash-course/classification/imbalanced-data

Topics

#Class Imbalance Handling#Data Preprocessing#Oversampling Techniques#Data Acquisition Strategy

Community Discussion

No community discussion yet for this question.

Sign up to join the discussion
Full MLS-C01 Practice