Glossario IA
Il dizionario completo dell'Intelligenza Artificiale
Simple Linear Regression
Statistical model that establishes a linear relationship between a single explanatory variable and a continuous target variable in the form Y = β₀ + β₁X + ε.
Multiple Linear Regression
Extension of linear regression using multiple explanatory variables to predict a continuous target variable according to Y = β₀ + ΣβᵢXᵢ + ε.
Coefficient of Determination (R²)
Statistical metric ranging from 0 to 1 measuring the proportion of variance of the target variable explained by the regression model.
Ordinary Least Squares (OLS)
Method for estimating regression parameters by minimizing the sum of squared residuals between observed and predicted values.
Residuals
Differences between observed values and values predicted by the regression model, representing prediction errors.
Multicollinearity
Phenomenon where multiple explanatory variables are highly correlated with each other, making coefficient estimation unstable.
Heteroscedasticity
Violation of the homoscedasticity assumption where the variance of residuals is not constant across the levels of explanatory variables.
Influential Values
Observations that, if removed from the dataset, would cause substantial changes in coefficient estimates.
Cook's Distance
Statistical measure that quantifies the influence of an individual observation on the predicted values of a regression model.
Ridge Regression
Regularization technique that adds an L2 penalty to coefficients to reduce variance and address multicollinearity.
LASSO Regression
Regularization method using an L1 penalty that can shrink some coefficients to exactly zero, performing variable selection.
Bias-Variance Tradeoff
Fundamental trade-off between bias error (oversimplification) and variance error (overfitting to training data).
Gradient Descent
Iterative optimization algorithm that adjusts coefficients to minimize the cost function by moving in the direction of the negative gradient.
Prediction Interval
Range of likely values containing the future individual observation with a specified confidence level, wider than a confidence interval.
Significance Test
Statistical test evaluating whether a regression coefficient differs significantly from zero, typically using the t-statistic.