prediction

Prediction is the act of using observed data and prior knowledge to infer the likely value of a future or unknown quantity. In statistics and data science, prediction emphasizes estimating outcomes for new cases rather than explaining why those outcomes occur. A prediction model combines inputs (features) to produce an estimate of a target variable, typically learned from historical data and evaluated on separate data to assess performance.

Predictions can be point estimates, which specify a single value, or probabilistic predictions, which provide a distribution over possible values. Related concepts include prediction intervals, which give a range expected to contain the true value with a specified probability, and confidence or credible intervals that communicate associated uncertainty.

Common methods range from traditional statistics (regression, time series analysis) to modern machine learning (neural networks, ensemble methods, decision trees). In domains that require explicit uncertainty quantification, probabilistic and Bayesian approaches are often employed. Evaluation uses metrics such as RMSE or MAE for point predictions, and Brier score, log loss, or calibration measures for probabilistic predictions.

Applications span many fields, including weather forecasting, economics and finance, epidemiology, engineering, and sports analytics. Limitations and challenges include data quality and leakage, nonstationarity, model misspecification, and overfitting. Predictions reflect uncertainty and do not imply causation; distinguishing predictive accuracy from causal understanding remains essential. In time-series contexts, forecasting is a specialized form of prediction that emphasizes future values in sequential data.