Forecasting with Auto ARIMA in Python

ARIMA is a popular and powerful statistical time series forecasting method.

Photo by Jack B on Unsplash

ARIMA Overview

ARIMA stands for (A)uto-(R)egressive (I)ntegrated (M)oving (A)verages. Here is a breakdown of what each of these terms mean:

• Auto Regressive: This essentially means ARIMA looks back at historical data to help predict the next data point. Think of this as a lag term in the model (how far back are we looking to help predict the next data point). In ARIMA, the “AR” term is also referred to as ‘p’ and can be manually set by looking at PACF plots.
• Integrated: This term in ARIMA refers to the differencing of values in a time series dataset. This helps transform the series into a stationary series. When a time series is stationary, the mean stays approximately the same. There is no trend upward or downward over time. In ARIMA, the “I” is also referred to as ‘d’. An ADF test can help determine if a series is stationary.
• Moving Averages: The Moving Averages component of ARIMA is essentially a window. This component looks at previous values to help gauge the trend of the series. In ARIMA, this term is referred to as ‘q’ and can manually be set by looking at ACF plots.