ARIMA模型是通过将预测对象随时间推移而形成的数据序列当成一个随机序列，进而用一定的数学模型来近似表述该序列。根据原序列是否平稳以及回归中所包含部分的不同分为AR、MA、ARMA以及ARIMA过程。
在模型的使用过程中需要根据时间序列的自相关函数、偏自相关函数等对序列的平稳性进行判别；而对于非平稳序列一般都需要通过差分处理将其转换成平稳序列（ARIMA）；对得到的平稳序列进行建模以确定最佳模型（AR、MA、ARMA或者ARIMA）。在使用中最重要也是最关键的就是对序列进行参数估计，以检验其是否具有统计意义。(The ARIMA model uses a mathematical model to approximate the sequence of data by forming a sequence of data that is predicted over time. It is divided into AR, MA, ARMA and ARIMA processes according to the stability of the original sequence and the included part of the regression.
In the process of the model according to the autocorrelation function, the partial sequence of stationary sequence autocorrelation function of discrimination; and for non stationary sequences generally need treatment to convert it into stationary sequence by difference (ARIMA); for the stationary sequences obtained were modeled to determine the best model (AR ARMA, MA, or ARIMA). In use, the most important and most important is to estimate the parameters of the sequence to test whether it is statistically significant.)