Choosing Between Classification and Regression
Quick answer
Will the customer churn, or how much will they spend next month? One is classification, the other regression — and picking the wrong framing dooms the project.
What is the difference?
Classification predicts a category: yes/no, spam/ham, one of N classes. Regression predicts a number: price, temperature, revenue. The question "which type of problem is this?" sounds trivial, but real projects routinely get it wrong — and the wrong framing wastes weeks.
A quick test: if the answer could be listed as distinct labels, it is classification. If the answer lives on a continuum and being off by 10% matters, it is regression.
Where teams get it wrong
Predicting churn risk as a probability (regression-style) and then choosing a threshold is usually better than hard-classifying churners: the probability feeds retention budgets that differ 10x per customer. Conversely, predicting exact next-month revenue (regression) when the business only acts on three buckets — low, medium, high — is noise amplification; classify instead.
Watch for targets that are 99% one class. Accuracy is meaningless there; look at precision, recall, and the confusion matrix. A model that always says "not fraudulent" is 99.8% accurate and 100% useless.
Same algorithms, different outputs
Most algorithms come in both flavors: logistic regression and k-NN classify, linear regression and k-NN regress. Decision trees do both natively. The evaluation changes more than the algorithm: accuracy, F1, and ROC-AUC for classification; RMSE and mean absolute error for regression.
In the Playground, the algorithm picker splits by task type so you cannot accidentally train a classifier on a continuous target — a small guardrail that saves a common rookie mistake.
Cite this page
AIalgo.si Editorial (2025). Choosing Between Classification and Regression. AIalgo.si Resources. https://aialgo.si/resources/classification-vs-regression