Thuật ngữ AI
Từ điển đầy đủ về Trí tuệ nhân tạo
Probabilistic Forecasting
Predictive approach generating a complete probability distribution for each time horizon rather than a simple point value.
Predictive Distribution
Complete probability distribution describing the uncertainty associated with future values of a time series.
Predictive Quantile
Value dividing the predictive distribution into specified proportions, allowing characterization of extremes and dispersion.
Temporal Bootstrap
Resampling method preserving the temporal structure to estimate forecast uncertainty.
Ensemble Forecasting
Technique combining multiple predictive models to generate an empirical distribution of forecasts.
Bayesian Model
Probabilistic framework incorporating prior knowledge and observations to produce predictive posterior distributions.
Quantile Forecasting
Direct generation of predictive quantiles without going through modeling the complete distribution.
Pinball Loss
Asymmetric loss function optimized for direct estimation of predictive quantiles.
Calibration of Forecasts
Evaluation of the consistency between assigned predictive probabilities and observed frequencies.
Reliability Diagram
Graphical tool evaluating calibration by comparing theoretical and empirical coverage rates.
Proper Scoring Score
Evaluation metric rewarding honest and well-calibrated probabilistic forecasts.
CRPS (Continuous Ranked Probability Score)
Evaluation score measuring the divergence between the predictive distribution and the actual observation.
Predictive Distribution Function
Cumulative function describing the probability that the future value is below a given threshold.
Distribution Forecast
Extension of probabilistic forecasting where the entire future distribution is modeled as a functional object.