wunder beta

📘 How do classifiers draw the line?

Your inbox this morning is a classification problem. The filter dumped your landlord's rent reminder into Spam and left two phishing notes sitting pretty in Primary. A human clerk in a 1936 Travelling Post Office would have made the same ki

4
lessons
~20 min
to learn
Adults
level
Start the course →

What you’ll learn

  1. Classification Fundamentals and Decision BoundariesDefine supervised classification against regression, name the parts of a labeled example, and distinguish underfit, a usable fit, and overfit on held-out data.Classification predicts a discrete label from features with known training labels. ISL contrasts that with regression, which predicts a continuous value. A classifier partitions feature space; the surfaces between regions are decision boundaries. The aim is generalization, which you diagnose on data the model has not seen.
  2. Core Classification Algorithms ComparedExplain, without fitting code, what logistic regression, k-NN, decision trees, naive Bayes, and SVMs each assume and do.Logistic regression applies a sigmoid to a linear score and draws a linear boundary. k-NN votes the neighborhood and degrades as dimensions grow. Trees split on thresholds and overfit alone. Naive Bayes multiplies feature likelihoods under a conditional-independence assumption that still works on text. An SVM maximizes the margin; kernels bend the cut; classic solvers get expensive as n grows.
  3. Confusion Matrix: Precision, Recall, and F1Keep train, validation, and test in their roles, read a confusion matrix, and compute accuracy, precision, recall, and F1 from the four counts.The test set is used once. Class imbalance lets a majority-class dummy score high accuracy while catching nothing. Fawcett's 2-by-2 — TP, FP, TN, FN — is the source of accuracy, precision, recall, and F1. Choose the metric from the cost of a miss versus a false alarm.
  4. Thresholds, ROC, AUC, and Metric ChoiceTreat the classification threshold as a deployment choice, read an ROC curve and its AUC as Fawcett defines them, and critique an imbalanced accuracy claim.The threshold trades recall against precision. ROC plots TPR against FPR across every threshold. AUC is the probability a random positive outranks a random negative (1.0 perfect, 0.5 chance). When positives are rare, a precision-recall curve is often clearer. An accuracy badge without a matrix is not a result.

Questions this course answers

What primarily distinguishes a classification task from a regression task?

Both are supervised and use labeled data. ISL's split is the type of target: qualitative categories for classification, a quantitative value for regression. Classification can be binary or multiclass.

A model fits the training data almost perfectly but performs poorly on new data. This is best described as:

Overfitting means the model captured noise and idiosyncrasies of the training set rather than generalizable structure. That is why you evaluate on held-out data.

What is a decision boundary?

A classifier partitions feature space into class regions. The surfaces between those regions are decision boundaries. Their shape depends on the algorithm.

Which statement about logistic regression is correct?

Logistic regression models class probability as a sigmoid of a linear combination of features, which yields a linear decision boundary in the original features. The last option describes k-NN.

What is the 'naive' assumption in naive Bayes?

scikit-learn's definition: naive Bayes applies Bayes' theorem with the assumption that features are conditionally independent given the class. That is often false and still useful, especially for text.

What does a support vector machine try to maximize?

An SVM seeks the maximum-margin separating boundary. The closest points that define that margin are the support vectors. Kernels allow nonlinear boundaries.

Grounded in trusted sources

  • [object Object]
  • [object Object]
  • [object Object]
  • [object Object]
  • [object Object]
  • [object Object]
  • [object Object]
  • [object Object]

Every Wunder lesson is built from real, reputable sources — never invented.

Related courses

📘How do you validate a model honestly?📘How does a neural network actually learn?📘How do machines read language?📘Keep transformer from meaning RNN clone📘Keep fine-tuning from meaning scratch📘Keep prompt from meaning magic spell📘How do you evaluate an AI product in the wild?📘How do humans stay in an AI loop?

Wunder is a personalized learn-anything platform — tell it any topic and it builds a beautiful, fact-checked course in minutes, with narration, a knowledge check, and a college-style University track.

All topics · Home

© 2026 Wunder Learning LLC · Terms & Privacy