ARIMA models use past patterns in time series data to forecast future values of a variable when causal factors are unknown. The Box-Jenkins methodology identifies the best ARIMA model by examining a time series' characteristics to transform it into white noise. ARIMA models include moving average (MA), autoregressive (AR), and mixed (ARMA) components. Stationarity is important and differencing can make data stationary. The steps are to identify the model type, estimate parameters, diagnose the model fit, and forecast.