A telematics dataset arrives with a driver's age, home state, a policy ID string, and a claim-count column. Before you fit anything, you must name each variable's type. Get the naming wrong and every downstream choice (chart, transform, model) inherits the error.
Target versus predictor. Every supervised problem splits variables into two roles. The target variable (also called the response, dependent, or outcome variable) is what you want to predict. The predictor variables (also called features, inputs, covariates, or independent variables) are what you use to predict it. In a claim model, claim cost is the target; driver age, vehicle type, and territory are predictors.
KEY: The role is defined by the business question, not by the data type. The same column can be a target in one model and a predictor in another. If you are predicting it, it is the target.
Types of values. A variable's value type decides which charts and transforms are legal.
Continuous variables can, in principle, take infinitely many values between any two points. Premium of $1,204.37 and loss of $8,150.00 are continuous.
Common mistakes
- Treating a numeric code as continuous. ZIP, policy ID, and territory code are labels. Averaging them or feeding them raw as numbers produces garbage; store them as factors.
- Confusing discrete with categorical. A claim count of 0, 1, 2 is discrete numeric; its mean (1.7) is valid. A factor like state has no numeric mean at all.
- Encoding an ordinal as nominal. Dropping the order in safety rating (1 to 5) discards real information the model could use to rank risk.
Bottom line
- The target (response) is the outcome you predict; predictors (features) are the inputs.
- Continuous variables take any value in a range; discrete variables take countable, separated values.
- Categorical (factor) variables hold a finite set of levels; string variables hold free text.
- A factor is ordered when its levels have a meaningful sequence (low, medium, high).
Exam shortcut
Decide role first, then value type, then ordering: ask "am I predicting it?" (target vs predictor), "does it measure or label?" (numeric vs categorical), and "do the levels rank?" (ordinal vs nominal). Whenever a column is numeric, sanity-check whether it measures a quantity or just codes a category; codes like ZIP and ID are factors no matter how they are stored.
The full lesson (about 1,681 words, 11 min read) adds 2 worked examples, all 6 common mistakes, a self-check, free in the app.
Learning objectives
- 2b
Browse all free Exam PA lessons or jump into free Exam PA practice questions.