Building on research on dynamical systems, neural networks, dynamic Bayesian networks, and time series analysis methods, the Dynamic Forecasting System is a next generation forecasting engine that delivers unparallelled forecasting accuracy.

Using the results of data preprocessing and feature selection, the Dynamic Forecasting System "learns" to forecast data by repeatedly processing historical data and automatically discovering inherent patterns in the data to be predicted, exploiting latent temporal relationships, and nonlinear relationships with selected external input data (features) such as customer or traffic characteristics, weather, holidays, economic indicators.

The Dynamic Forecasting System can also produce confidence intervals on predictions, to be used in forecasting application where knowing upper and lower bounds of likely values is important. Forecasts of most likely values can be generated, as well as complete probability distributions over possible values, providing useful information about the uncertainty of outcomes, such as the uncertainty in individual travel time forecasts.