Cox Model Nugget

Cox regression models represent the equations estimated by Cox nodes. They contain all of the information captured by the model, as well as information about the model structure and performance.

When you run a stream containing a generated Cox regression model, the node adds two new fields containing the model's prediction and the associated probability. The names of the new fields are derived from the name of the output field being predicted, prefixed with $C- for the predicted category and $CP- for the associated probability, suffixed with the number of the future time interval or the name of the time field that defines the time interval. For example, for an output field named churn and two future time intervals defined at regular intervals, the new fields would be named $C-churn-1, $CP-churn-1, $C-churn-2, and $CP-churn-2. If future times are defined with a time field tenure, the new fields would be $C-churn_tenure and $CP-churn_tenure.

If you have selected the Append all probabilities settings option in the Cox node, two additional fields will be added for each future time, containing the probabilities of survival and failure for each record. These additional fields are named based on the name of the output field, prefixed by $CP-<false value>- for the probability of survival and $CP-<true value>- for the probability the event has occurred, suffixed with the number of the future time interval. For example, for an output field where the "false" value is 0 and the "true" value is 1, and two future time intervals defined at regular intervals, the new fields would be named $CP-0-1, $CP-1-1, $CP-0-2, and $CP-1-2. If future times are defined with a single time field tenure, the new fields would be $CP-0-1 and $CP-1-1, since there is a single future interval

If you have selected the Calculate cumulative hazard function settings option in the Cox Node, an additional field will be added for each future time, containing the cumulative hazard function for each record. These additional fields are named based on the name of the output field, prefixed by $CH-, suffixed with the number of the future time interval or the name of the time field that defines the time interval. For example, for an output field named churn and two future time intervals defined at regular intervals, the new fields would be named $CH-churn-1 and $CH-churn-2. If future times are defined with a time field tenure, the new field would be $CH-churn-1.