Glossario IA
Il dizionario completo dell'Intelligenza Artificiale
Lag Feature
Variable created by shifting a time series by one or more time periods to capture temporal dependencies and memory effects in the data.
Moving Average
Statistical calculation that analyzes data points by creating a series of averages from different subsets of the complete data series to smooth out short-term fluctuations.
Seasonal Decomposition
Statistical method separating a time series into three fundamental components: trend, seasonality, and residual to better understand and model temporal patterns.
Temporal Feature Engineering
Process of creating predictive variables from raw temporal data to improve machine learning model performance by capturing temporal patterns.
Differencing
Transformation technique involving calculating the differences between consecutive observations to make a time series stationary by eliminating trend and seasonality.
Sliding Window
Analytical approach applying a statistical function to a fixed subset of data that progressively moves along a time series to capture local patterns.
Additive Seasonality
Type of decomposition where seasonal variations are approximately constant over time and add to the trend: Y(t) = Trend + Seasonality + Residual.
Multiplicative Seasonality
Decomposition model where seasonal variations vary proportionally to the trend: Y(t) = Trend × Seasonality × Residual.
Trend Feature
Variable characterizing the direction and intensity of long-term change in a time series, often represented by temporal polynomials or long moving averages.
Stationarization
Process of transforming time data to achieve stationarity, where statistical properties like mean and variance remain constant over time.
Window Function
Function applied to a specific time window to calculate aggregate statistics such as sum, mean, standard deviation, or minimum/maximum over that period.
Temporal Feature Scaling
Normalization technique for temporal features that considers chronological order and temporal dependencies to avoid future information leakage.
Holiday Effect Features
Binary or continuous variables indicating the presence or proximity of holidays and special events to capture their impacts on time series.