Deep Learning (DL) · Intermediate level

Intermediate level · Module 5: Evaluating Deep Learning Models · Lesson 2 of 3

Error Analysis

Look systematically at what the model gets wrong and why, to guide improvements.

In this lesson you will

  • Categorise errors by cause
  • Prioritise fixes by impact

Key concepts

error categoriesconfusion pairsconfidently wrong predictionsfix priorities

The lesson

Collect misclassified examples and sort them into categories: blur, glare, partly covered package, similar-looking products, label error. Count each category.

The confusion matrix shows which pairs of classes are confused; confidently wrong predictions often reveal label errors or missing training conditions.

Fix the largest categories first: more training photos with glare, better labels or merging truly indistinguishable classes.

Step by step

  1. Collect 50 misclassified examples.
  2. Categorise them by cause.
  3. Count categories and confusion pairs.
  4. Propose fixes for the top two.

Practice activity

Write the error analysis for the project.

Expected result: Categorised, counted errors with prioritised fixes.

Check your understanding

  1. Why categorise errors?

    Show answer

    To find which problems are most common and fix them first.

  2. What can confidently wrong predictions reveal?

    Show answer

    Label errors or conditions missing from training data.

Taught live by an instructor — online or in person. Free course lessons are for self-study and do not lead to a certificate — see registration options.