Time series forecasting algorithms

A wide variety of time series forecasting methods are available today for both univariate and multivariate forecasting. Some of them are only effective on univariate time series, while others are suitable for causal forecasting. Time series forecasting algorithms can be grouped into two categories:

• Classic algorithms are traditional time series forecasting algorithms that work only on univariate time series. These algorithms have been used for many years for time series forecasting.
• Machine learning algorithms can be used for univariate and multivariate forecasting.

Mixed type algorithms, a third category, combine multiple algorithms of classic and machine learning types. The following table shows the categorization of the available algorithms in Demand Guru:

Classic	Machine Learning	Mix
Naïve Forecasting	Arimax	Tournament
Simple Moving Average	Support Vector Machine (SVM)	Ensemble
Exponential Smoothing	Quantile Random Forest (QRF)	Automatic
Arima	Neural Networks
Auto Arima	Stochastic Gradient Boosting / Gradient Boosting Machine (GBM)

To determine which algorithm works best for a specific demand model, you must understand the logic behind each one.