Explain ML basics: imbalance, metrics, bias-variance

Q: Explain ML basics: imbalance, metrics, bias-variance

This is a Machine Learning interview question from Amazon for Machine Learning Engineer roles. View the full question and solution on PracHub.

Q: How do I approach Machine Learning interview questions?

Machine Learning questions require understanding of core concepts and practice. PracHub provides solutions with explanations to help you master machine learning interviews.

Question

Handling Class Imbalance, Bias–Variance, Metrics, and Model Choices

Context

You are building a supervised classifier for a highly imbalanced task (e.g., fraud detection) where the positive class is rare. Discuss how you would:

Diagnose class imbalance.
Address imbalance using data-level and algorithm-level techniques (e.g., over/under-sampling, SMOTE variants, class-weighted losses, focal loss), data augmentation, threshold tuning, and probability calibration.
Design validation to avoid leakage and to reflect real class priors.
Define the bias–variance trade-off and outline concrete steps for high bias vs. high variance (model capacity, regularization, features, data).
Choose appropriate evaluation metrics for imbalanced data and justify them—contrast accuracy, ROC-AUC, PR-AUC, F1/Fβ, recall@k/precision@k, and expected business cost; explain when each is preferable.
Briefly explain the core ideas and inductive biases of CNNs vs. Transformers and when you would prefer each for text, images, sequences, or tabular data.

Explain ML basics: imbalance, metrics, bias-variance

Handling Class Imbalance, Bias–Variance, Metrics, and Model Choices

Context

Solution (Locked)

Comments (0)