Binary vs. multi-class and when to use each
Classification predicts a discrete label for each input—text, images, audio, or any data you can describe in a prompt.
Type | Example | Output | Use Cases |
---|---|---|---|
Binary | spam / not-spam, approve / reject | Single label (2 options) | Content moderation, fraud detection, A/B testing |
Multi-Class | news topic → Politics / Sports / Tech | Single label (3+ options) | Document routing, sentiment analysis, intent detection |
Multi-Label | article tags → AI + Healthcare + Ethics | Multiple labels | Product tagging, skill extraction, content categorization |
Domain | Task | Traditional ML Challenge | LLM Advantage |
---|---|---|---|
Customer Support | Intent classification | Requires large labeled dataset | Works with just intent descriptions |
Content Moderation | Toxicity detection | Struggles with context/sarcasm | Understands nuanced harmful content |
E-commerce | Product categorization | Needs product-specific training | Generalizes across product types |
HR/Recruiting | Resume screening | Biased on historical data | Can focus on skills vs. demographics |
Finance | Document classification | Regulatory compliance complexity | Adapts to new regulations quickly |
Healthcare | Symptom triage | Requires medical expertise | Leverages medical knowledge from training |
Need multiple labels per input? See our Multi-Label guide for handling non-exclusive categories.