Evaluation and confusion matrix in Computer Vision - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is a confusion matrix in machine learning?

A confusion matrix is a table that shows the number of correct and incorrect predictions made by a classification model, broken down by each class. It helps us understand how well the model is performing.

beginner

What do the terms True Positive (TP), False Positive (FP), True Negative (TN), and False Negative (FN) mean?

TP: Model correctly predicts positive class.<br>FP: Model incorrectly predicts positive class.<br>TN: Model correctly predicts negative class.<br>FN: Model incorrectly predicts negative class.

beginner

How is accuracy calculated from a confusion matrix?

Accuracy = (TP + TN) / (TP + TN + FP + FN). It measures the overall correctness of the model's predictions.

intermediate

Why might accuracy be misleading in imbalanced datasets?

If one class is much more common, a model can predict that class all the time and get high accuracy but perform poorly on the rare class. Other metrics like precision and recall help in such cases.

intermediate

What are precision and recall, and why are they important?

Precision measures how many predicted positives are actually positive (TP / (TP + FP)).<br>Recall measures how many actual positives were correctly predicted (TP / (TP + FN)).<br>They help evaluate model performance especially when classes are imbalanced.

What does the diagonal of a confusion matrix represent?

A Correct predictions

B Incorrect predictions

C False positives

D False negatives

The diagonal entries show the number of correct predictions for each class.

Which metric tells you the proportion of actual positives correctly identified?

A Accuracy

B Precision

C Recall

D Specificity

Recall measures how many actual positive cases were correctly predicted.

If a model predicts all samples as negative in a dataset with 90% negatives, what will be the approximate accuracy?

A 90%

B 10%

C 50%

D 100%

Predicting all negatives correctly matches the 90% negative class, so accuracy is about 90%.

Which of these is NOT part of a confusion matrix?

A False Positive

B False Negative

C True Positive

D True Neutral

There is no 'True Neutral' in confusion matrix terminology.

Why is a confusion matrix useful in computer vision classification tasks?

A It shows how many pixels are in an image

B It helps understand model errors per class

C It speeds up training

D It reduces image size

Confusion matrix helps analyze which classes the model confuses, improving understanding of errors.

Explain how to interpret a confusion matrix and what insights it provides about a classification model.

Think about how the matrix shows where the model gets things right or wrong.

Describe why accuracy alone might not be enough to evaluate a model and which other metrics can help.

Consider cases where one class dominates the data.