Explain core ML concepts and metrics

You are in an interview for a Data Scientist position. Respond to the machine learning fundamentals questions below in a clear, concise way.

Concepts

1. Describe the bias–variance tradeoff. How is it connected to overfitting versus underfitting?
2. Name common regularization techniques (such as L1/L2 or early stopping). What issue does regularization address?

Imbalanced classification

3. When working with imbalanced datasets, which evaluation metrics do you favor, and why? Contrast accuracy, precision, recall, F1, PR-AUC, and ROC-AUC.
4. Give definitions and formulas for precision and recall. In what situations would you favor one over the other?

Logistic regression / probabilities

5. In logistic regression, what is the model’s raw output prior to probability conversion? How is it transformed into a probability?

ROC-AUC interpretation

6. What does a ROC-AUC of 0.8 indicate? Give an intuitive reading and at least one caveat.

Models

7. What are ensemble models, and why do they frequently beat a single model?
8. With tree-based models such as decision trees, random forests, or gradient boosting, list the important hyperparameters and explain their impact on bias and variance.

Output activations

9. Contrast sigmoid and softmax: when is each appropriate, and how are their outputs different?

Overview: This question assesses understanding of essential machine learning concepts and competencies, including the bias–variance tradeoff, regularization, evaluation metrics for imbalanced classification (accuracy, precision, recall, F1, PR-AUC, ROC-AUC), probabilities in logistic regression, ensemble methods, tree-based hyperparameters, and output activations (sigmoid vs. softmax). It is frequently used in technical interviews for Machine Learning and Data Scientist roles to evaluate reasoning about model behavior, metric choice, trade-offs, and interpretability, checking both conceptual knowledge and practical skill in model evaluation and tuning.