Which model only considers a single variable in its prediction?

The correct choice identifies the model that exclusively focuses on a single variable for its predictions. In the context of time-series forecasting, the Autoregressive Integrated Moving Average (ARIMA) can indeed model a single variable while incorporating aspects of the variable's own historical values as well as a moving average of the forecast errors. This model focuses on the temporal dynamics of one variable, making it suitable for univariate time series analysis.

In contrast, linear regression typically considers multiple independent variables when predicting a dependent variable. While it can be used in a univariate context, it is not limited to that. Support vector machines allow for multi-dimensional data and operate in a feature space that can involve numerous variables and their interactions. Random forests, on the other hand, utilize an ensemble of decision trees that take into account multiple predictors, enhancing predictive performance by aggregating various variables.

Thus, ARIMA is specifically designed to address single-variable time-series data accurately, making it the correct answer for the question posed.