← AI Terminology
Predictors
Predictors are the input variables used by a model to predict an outcome — synonymous with features or independent variables.
The term is more common in statistics and regression contexts, while "features" is the preferred term in machine learning.
The term is more common in statistics and regression contexts, while "features" is the preferred term in machine learning.
Why It Matters in AI
Naming matters for communication: statisticians say predictors, ML engineers say features, and data scientists say both. Understanding the equivalence helps you read across disciplines — a regression paper's "predictors" are the same thing as a neural network paper's "input features." Selecting good predictors is the central challenge of building any model, because no algorithm can compensate for uninformative inputs.
Key Points
| Aspect | Description |
|---|---|
| Also called | Features, input variables, independent variables, covariates, regressors, explanatory variables |
| Common tools | statsmodels (statistical framing), sklearn (ML framing), R (lm/glm use "predictors" natively) |
| Contrast with | Response / target / dependent variable — the thing being predicted |
| In regression | Each predictor gets a coefficient (e.g. β₁) measuring its linear relationship to the target |
| Multicollinearity | Highly correlated predictors cause instability in linear models; regularisation or PCA helps |
| Selection methods | Correlation, LASSO regularisation, recursive feature elimination, mutual information |
Simple Analogy
A weather forecast uses temperature, humidity, and wind speed to predict tomorrow's rain — each is a predictor. The forecast is only as good as these inputs; if the sensors are broken or you forget to include humidity, the prediction suffers no matter how good the forecasting algorithm is.
Common Usage Examples
lm(salary ~ age + education + experience, data=df)in R — each right-hand variable is a predictormodel.coef_in scikit-learn linear models gives a weight per predictor- Variance Inflation Factor (VIF) used to detect collinear predictors in regression diagnostics
- SHAP values rank predictors by their contribution to each individual prediction
- In time series: lagged values of the target (e.g. yesterday's price) used as predictors for today's
Summary
In short: "Predictors" is just the statistician's word for features — the inputs you give a model so it can make a prediction.