To do this,
- A model is developed from a training data set where the values of input and output are previously known
- The model generalizes the relationship between the input and output variables
- And uses this model to predict for the data set where only input variables are known