Classification

Classification is a machine learning task where a model assigns each input to one of a fixed set of categories. Fraud or not fraud. High risk, medium risk, or low risk. Which product department a customer inquiry belongs to. The model learns the pattern from labeled examples, then applies it to new inputs it hasn't seen before. Classification underpins a wide range of practical business applications: fraud detection, content moderation, document routing, customer intent recognition, and medical screening.

In classification, how the model fails matters as much as how often it succeeds. Blocking a legitimate transaction (false positive) and missing a fraud case (false negative) are both errors — but they have very different business consequences. A model optimized purely for overall accuracy on an imbalanced dataset often learns to predict the common outcome nearly every time, missing the rare cases entirely. Those rare cases are usually exactly what the model was built to catch. The precision-recall trade-off is ultimately a business decision, and it belongs to the business — not the data science team.