In order to build predictive models, training data is required that has examples of previous observed behaviour. The modelling software will then take this data and try to combine the 'inputs' in such as was as to model the 'output'. These models can then be used on new data to predict what the output is likely to be for a given set of inputs.

The way the 'combining' of the inputs is done depends on the mathematical algorithm used. Tiberius contains an ever growing suite of algorithms such as Neural Networks, Logistic Regression and Regression Splines.

There are two 'families' of predictive models,

**Classification** - predicting an event, such as 'will a customer respond',

**Regression** - predicting something that is continuous, such as tomorrows foreign exchange rates.

In *classification* models, the 'target' in the training data will be be one of two discrete value such as 1 or 0, whereas for regression it can take on any continuous value.

Tiberius has tools for each type of problem.

These sections will guide you through the process of how to actually build models without going through any details of the actual algorithms themselves.