DecodeAI
Sign inStart Learning
← Question Bank

Machine Learning

Performance Metrics

Interview questions on Performance Metrics.

48 questions

All subtopicsRegression MetricsClassification MetricsClustering MetricsMetrics in NLP

Classification Metrics

Q1. What metrics we can use to evaluate a classifier?

Sign in to bookmark

Classification Metrics

Q2. What is the purpose of a confusion matrix?

Sign in to bookmark

Classification Metrics

Q3. Explain True Positive (TP) and False Positive (FP).

Sign in to bookmark

Classification Metrics

Q4. Define True Negative (TN) and False Negative (FN).

Sign in to bookmark

Classification Metrics

Q5. How can you use a confusion matrix to choose an appropriate threshold for a binary classifier?

Sign in to bookmark

Classification Metrics

Q7. Write the mathematical expression of accuracy score?

Sign in to bookmark

Classification Metrics

Q9. How does accuracy is related with negative log(logistic) loss?

Sign in to bookmark

Classification Metrics

Q10. What is the main issue with accuracy score?

Sign in to bookmark

Classification Metrics

Q11. How can we overcome with accuracy score limitations?

Sign in to bookmark

Classification Metrics

Q12. What do you mean by precision? How can we calculate it?

Sign in to bookmark

Classification Metrics

Q14. What is the range of precision value?

Sign in to bookmark

Classification Metrics

Q15. What are the weaknesses of precision as a metric in classification, and how can it be misleading in certain scenarios?

Sign in to bookmark

Classification Metrics

Q16. State one scenario where you should use precision of the system as evaluation metric?

Sign in to bookmark

Classification Metrics

Q19. What does recall measures? How is it calculated?

Sign in to bookmark

Classification Metrics

Q20. What does higher value of recall indicates?

Sign in to bookmark

Classification Metrics

Q23. What is the main limitation of recall metric?

Sign in to bookmark

Classification Metrics

Q24. State one scenario where recall is more desirable?

Sign in to bookmark

Classification Metrics

Q25. Explain the tradeoff between recall and precision?

Sign in to bookmark

Classification Metrics

Q27. What is F-1 Score? When should we use it?

Sign in to bookmark

Classification Metrics

Q29. What are the weaknesses of F-1 score?

Sign in to bookmark

Classification Metrics

Q30. Explain the difference between Type I and Type II errors in the context of a confusion matrix.

Sign in to bookmark

Classification Metrics

Q31. What is the main limitations with confusion matrix summarization?

Sign in to bookmark

Classification Metrics

Q33. State the use of precision and recall curve?

Sign in to bookmark

Classification Metrics

Q34. How can we construct precision and recall curve?

Sign in to bookmark

Classification Metrics

Q35. What are the limitations of precision and recall curve?

Sign in to bookmark

Classification Metrics

Q37. For an ideal model what should ROC-AUC?

Sign in to bookmark

Classification Metrics

Q38. State what does mean to have AUC of 0.2?

Sign in to bookmark

Classification Metrics

Q39. Can we use ROC-AUC score for multi-class classification?

Sign in to bookmark

Classification Metrics

Q41. What is the limitations of ROC-AUC score?

Sign in to bookmark

Classification Metrics

Q42. Your team is building a system to aid doctors in predicting whether a patient has cancer or not from their X-ray scan. Your colleague announces that the problem is solved now that they’ve built a system that can predict with 99.99% accuracy. How would you respond to that claim?

Sign in to bookmark

Classification Metrics

Q43. Given a binary classifier that outputs the following confusion matrix.

Sign in to bookmark

[""Predicted TruePredicted FalseActual True3020Actual False540]\begin{bmatrix} \textbf{""} & \textbf{Predicted True} & \textbf{Predicted False} \\\\ \textbf{Actual True} & 30 & 20 \\\\ \textbf{Actual False} & 5 & 40 \\\\ \end{bmatrix}

  1. Calculate the model’s precision, recall, and F1.
  2. What can we do to improve the model’s performance?

Classification Metrics

Q44. Consider a classification where $99%$ of data belongs to class A and $1%$ of data belongs to class B.

Sign in to bookmark
  1. If your model predicts A 100% of the time, what would the F1 score be? Hint: The F1 score when A is mapped to 0 and B to 1 is different from the F1 score when A is mapped to 1 and B to 0.
  2. If we have a model that predicts A and B at a random (uniformly), what would the expected F1F_1 be?

Classification Metrics

Q45. Show that the negative log-likelihood and cross-entropy are the same for binary classification tasks.

Sign in to bookmark

Classification Metrics

Q46. For classification tasks with more than two labels (e.g. MNIST with $10$ labels), why is cross-entropy a better loss function than MSE?

Sign in to bookmark

Classification Metrics

Q47. Consider a language with an alphabet of $27$ characters. What would be the maximal entropy of this language?

Sign in to bookmark

Classification Metrics

Q48. Suppose you want to build a model to predict the price of a stock in the next 8 hours and that the predicted price should never be off more than $10%$ from the actual price. Which metric would you use?

Sign in to bookmark