Classification vs Regression: Choose the Right Target

ML basics • 5-7 min read • 15-30 min to practice

TL;DR: Classification predicts discrete labels; regression predicts continuous values. Pick models and metrics that match your target and business cost of errors.

Definitions

• Classification: predict categories (spam/not spam, churn/no churn).
• Regression: predict continuous values (price, demand, temperature).

Metric guidance

• Classification: accuracy, precision, recall, F1, AUC-ROC. Use precision/recall when classes are imbalanced.
• Regression: MAE, RMSE, MAPE. Use MAPE when percent error is meaningful.

Choosing algorithms

• Small data, interpretable: logistic regression (classification), linear regression (regression).
• Complex relationships: tree ensembles, neural networks.

Example: When to transform

• Turn regression into classification when thresholds drive action (e.g., convert predicted spend into “high value” bucket).
• Beware information loss when binning continuous targets.

Quick experiments

• Always compare to a baseline: majority class or mean prediction.
• Use calibration plots for probabilistic classifiers.

Next steps

The next article shows common evaluation metrics and practical examples for computing them.

Visual: classification vs regression

Replace with a small SVG illustrating discrete vs continuous targets and example metrics.