Explain core ML concepts and metrics
Overview
This question evaluates understanding of core machine learning concepts and competencies such as the bias–variance tradeoff, regularization, evaluation metrics for imbalanced classification (accuracy, precision, recall, F1, PR-AUC, ROC-AUC), logistic regression probabilities, ensemble methods, tree hyperparameters, and output activations (sigmoid vs softmax). It is commonly asked in technical interviews for Machine Learning and Data Scientist roles to assess reasoning about model behavior, metric selection, trade-offs and interpretability, testing both conceptual understanding and practical application of model evaluation and tuning.
You are interviewing for a Data Scientist role. Answer the following ML fundamentals questions clearly and concisely.
Concepts
- Explain the bias–variance tradeoff . How does it relate to overfitting vs. underfitting ?
- What are common forms of regularization (e.g., L1/L2, early stopping)? What problem does regularization solve?
Imbalanced classification
- For imbalanced datasets , which evaluation metrics do you prefer and why? Compare accuracy, precision, recall, F1, PR-AUC, ROC-AUC .
- Define precision and recall and provide their formulas. When would you optimize for one over the other?
Logistic regression / probabilities
- In logistic regression , what is the model’s raw output before converting to a probability? How is it mapped to a probability?
ROC-AUC interpretation
- What does it mean if ROC-AUC = 0.8 ? Provide an intuitive interpretation and at least one caveat.
Models
- What are ensemble models and why do they often outperform a single model?
- For tree-based models (decision trees / random forests / gradient boosting), name key hyperparameters and describe how they affect bias/variance.
Output activations
- Compare sigmoid vs. softmax : when do you use each, and how do their outputs differ?
Solution
Login required