Main Concept
Classification is a supervised learning technique used to predict categorical labels — assigning input data to one or more discrete categories or classes. The output is not a number, but a category.
Unlike regression (which predicts “how much?”), classification answers “what is it?” or “which category does it belong to?”
How It Works
You train a model on labeled data where each example belongs to a known category. Once trained, when you give the model a new input, it predicts which category it belongs to based on what it learned.
Example: You train a spam filter on emails labeled “spam” and “not spam.” The model learns what characteristics make an email spam. When a new email arrives, the model classifies it as spam or not spam.
Three Types of Classification
1. Binary Classification
Predicts between exactly two categories.
Examples:
- Spam or not spam (emails)
- Fraudulent or legitimate (transactions)
- Disease or healthy (medical diagnosis)
- Yes or no (any yes/no question)
2. Multi-class Classification
Predicts between three or more categories, but each example belongs to exactly ONE category.
Examples:
- Animal type: dog, cat, or giraffe
- Image recognition: cat, dog, bird, or rabbit
- Disease classification: disease A, disease B, disease C, or healthy
- Document type: report, email, news, or social media
3. Multi-label Classification
Each example can belong to MULTIPLE categories at once.
Examples:
- Movie genres: a movie can be both “action” AND “comedy”
- Document tags: an article can be tagged “technology” AND “business” AND “finance”
- Symptom diagnosis: a patient can have multiple diseases simultaneously
Key Characteristics
- Output is categorical — a specific class or label, not a number
- Requires labeled data — you must know which category each training example belongs to
- Discrete decisions — unlike regression (which is continuous), you’re picking from a specific set of options
- Clear boundaries — the model draws boundaries between categories
Common Classification Examples
| Problem | Input | Output | Type |
|---|---|---|---|
| Email spam detection | Email content | Spam / Not spam | Binary |
| Image recognition | Image | Cat / Dog / Bird | Multi-class |
| Disease diagnosis | Medical tests | Disease A / B / C / Healthy | Multi-class |
| Movie recommendations | Movie features | Action, Comedy, Drama (multiple labels possible) | Multi-label |
| Credit fraud detection | Transaction data | Fraudulent / Legitimate | Binary |
| Customer churn | Customer behavior | Will leave / Will stay | Binary |
Classification vs. Regression
| Aspect | Classification | Regression |
|---|---|---|
| Output type | Discrete category | Continuous number |
| Example | ”This is a dog" | "This dog weighs 20 kg” |
| Question it answers | ”What is it?” or “Which category?" | "How much?” or “How many?” |
| Data pattern | Points grouped by category | Points scattered around a line |
Algorithms for Classification
Common algorithms used for classification (exam won’t ask you to code these):
- Logistic Regression — simple, interpretable
- Decision Trees — easy to understand
- SVM (Support Vector Machine) — powerful for complex problems
- k-NN (k-Nearest Neighbors) — finds similar examples in training data
- Neural Networks — learns complex patterns
- Random Forests — ensemble of decision trees
Why Classification Matters
Classification is used for decision-making, security, and automation:
- Security: spam detection, fraud detection, intrusion detection
- Healthcare: disease diagnosis, patient risk assessment
- Business: customer churn prediction, sentiment analysis
- Automation: image recognition, voice recognition, document classification
AIF-C01 Exam Relevance
The exam expects you to:
- Recognize classification problems (predicting a category or label)
- Know that classification requires labeled training data
- Understand the difference between binary, multi-class, and multi-label
- Distinguish it from regression
- Know that different algorithms can be used for classification
Exam tip: If you see “classify, categorize, predict a label, or identify which category,” it’s a classification problem. If the output is a specific category (not a number), it’s classification.