Classification
An ML task that predicts a discrete category — spam vs not-spam, churn vs no-churn.
What is Classification?
Classification problems have a finite set of possible outputs (classes). Binary classification has two classes (yes / no, fraud / not fraud). Multi-class has more (low risk / medium risk / high risk). Multi-label allows multiple classes per input simultaneously.
Common algorithms: logistic regression, decision trees, random forests, XGBoost, neural networks. The choice depends on data size, interpretability needs, and accuracy requirements.
Evaluation goes beyond accuracy — you need to consider **precision** (when you predict positive, how often are you right?), **recall** (of all real positives, how many did you catch?), **F1 score**, **ROC-AUC**. Picking the right metric per business problem is half the work.
Classification is the #1 ML pattern in production. Every fraud detector, churn predictor, spam filter, and credit decision is a classifier.
A Pune logistics startup classifies every delivery request into low / medium / high difficulty using 50 features (distance, weather, neighbourhood, time-of-day). The classification routes orders to the right driver pool — reducing failed deliveries 18%.
Want to master this?
Learn Classification in a structured cohort
3-month live program with mentors, real projects, and 50+ partner placement support.