DataTerminology.comNEW
Also known as: Bias Term, Intercept
The additive parameter in linear models representing the predicted value when all input features are zero.
DESCRIPTION
In regression and many machine learning models, the bias (intercept) shifts the model’s output independently of the input features. It captures baseline effects not explained by feature weights and is crucial when inputs are centered away from zero. In classification, it moves the decision boundary. Many solvers compute the intercept unless explicitly disabled (fit_intercept=False). Intercepts are commonly excluded from regularization to avoid biasing the baseline. Proper feature scaling and centering can stabilize intercept estimation.
ALTERNATIVE DEFINITIONS
NIST/SEMATECH e-Handbook of Statistical Methods
The intercept is the value of the response when the predictor(s) is (are) zero.
Defines the intercept operationally as the response at zero predictors in regression.
scikit-learn
fit_intercept bool, default=True Whether to calculate the intercept for this model. If set to False, no intercept will be used in calculations (i.e. data is expected to be centered).
Highlights the practical computation and centering implications of including an intercept.
Penn State STAT 501
The intercept, β0, is the mean value of the response when all predictors equal zero.
States the parameter meaning in the linear model with statistical notation.
RELATED TERMS
Linear Model Logistic Regression Gradient Descent Regularization Model Overfitting Model Underfitting Feature Scaling Decision Boundary
COMPETING TERMS
No-intercept Model
RELATIONS
- Broader: Bias
KNOWLEDGE AREAS
TARGET AUDIENCE
Data Scientist AI Engineer MLOps Engineer Data Engineer Data Modeler
EXTERNAL SOURCES
- Wikipedia - Intercept (statistics)
- NIST - NIST regression basics
- scikit-learn - LinearRegression API intercept_
- Stanford CS229 - Linear regression notes with intercept
TOP QUESTIONS
Why is the Bias (Mathematical) term important in linear models?
It captures the baseline level of the target when features are zero, ensuring the model can fit data not centered at the origin. Without it, predictions can be systematically shifted.
Category: Relevancy
How does the Bias (Mathematical) affect decision boundaries?
In linear classifiers, the intercept shifts the decision boundary parallel to itself, enabling better separation when classes are not centered around the origin.
Category: Example
Should the Bias (Mathematical) be regularized?
Common practice is not to regularize the intercept because penalizing it can introduce unintended baseline bias. Many libraries exclude it from penalties by default.
Category: BestPractice
When would you use a no-intercept model instead of including Bias (Mathematical)?
Use no-intercept when theory dictates the response must be zero at zero input (e.g., certain physical laws) or when data are mean-centered and design ensures it.
Category: Opinion
How does feature scaling impact the Bias (Mathematical) value?
Centering features (subtracting the mean) usually reduces the magnitude of the intercept, improving numerical stability and interpretability.
Category: Requirements
What is a common mistake regarding Bias (Mathematical)?
Disabling the intercept without justification can force the model through the origin, causing systematic errors and poor fit.
Category: Mistakes
How is Bias (Mathematical) represented in software libraries?
It often appears as an intercept_ attribute or a column of ones in the design matrix. Libraries provide a fit_intercept option to control its inclusion.
Category: Tooling
Does Bias (Mathematical) matter in neural networks?
Yes, neuron bias terms shift activation thresholds, enabling networks to model functions not passing through the origin.
Category: Relevancy
What is an intuitive analogy for Bias (Mathematical)?
Think of it as raising or lowering a straight line on a graph without changing its tilt.
Category: Example