Support Vector Machine

Support Vector Machine (SVM) is a supervised machine learning algorithm. These types of models are known as support vector regression (SVR) when used in a regression context.

The goal of support vector regression is to fit as many instances as possible between the margin while limiting margin violations. This algorithm forms a “tube” around the time series data by vectors (called support vectors) and evaluates the quality of the fit by a loss function (referred to as an epsilon loss function). The value of epsilon defines a margin of tolerance where no penalty is given. The boundaries of this tube are determined by mining the loss function.

The algorithm performs the “kernel” trick for non-linear regression. This operation allows it to operate in a high-dimensional feature space without ever computing the coordinates of the data in that space. The kernel trick significantly improves computational efficiency.

Support vector regressions are sensitive to feature scaling. Machine learning algorithms that consider distances between the observations must normalize the range of features.

If the demand data is random and/or does not exhibit seasonality and trend patterns, or if the time series length is too short, predictions by this algorithm can stray in the wrong direction because of overfitting.

Use this algorithm when:

• The length of time series is long (greater than 75 points).

• Causal data is available.

• Multiple user defined causal variables exist and you are using causal variables from the Trend Cloud.