A power company is building a model that predicts the next day's electricity usage using temperature, day of week, and past usage records as features, to use for generation planning. Rather than judging usage with categories such as "high / low", it must output a concrete quantity that can be used directly in the plan. Which task type is MOST appropriate for this?
A question on choosing the task type that predicts a continuous value.
Regression
Correct. Regression is a supervised learning task that predicts a continuous quantity (usage, sales amount, temperature, etc.). This scenario requires 'a concrete quantity that can be used directly in the plan' — that is, a continuous-value output — so regression applies.
Classification
Classification is a task that assigns inputs to discrete categories such as 'high / low'.
Here a category judgment is not enough; the requirement explicitly states the output must be a concrete quantity that can be used directly in the plan, so classification does not meet it.
Clustering
Clustering is an unsupervised learning task that groups unlabeled data into similar groups.
This scenario is supervised learning that predicts future values using known answers (past usage records), so grouping does not meet the requirement.
Anomaly detection
Anomaly detection is a task that finds data that differs greatly from normal (outliers, fraud, etc.).
This scenario predicts the normal usage itself to use in planning, so detecting anomalies does not meet the requirement.
Distinguish task types by the output.
・Regression: predicts a continuous value (amount, usage, temperature).
・Classification: predicts a category (high/low, spam/not spam).
・Clustering: groups data without labels (unsupervised).
Even for the same prediction, the requirement of 'is a category enough, or is the quantity itself required' separates classification from regression.