Types of predictive models

Predictive models are optimized to predict different types of outcome.

Scoring models

Scoring models predict a binary outcome, such as good or bad creditworthiness, high or low churn risk. Scoring models return a value known as the score, which places a case on a numerical scale. Typically, the range of scores is divided into intervals of increasing likelihood of one of two types of behavior, based on the behavior of the cases in the development sample that fall into each interval. High scores are associated with good performance and low scores are associated with bad performance.

The outcome of scoring models contains values that identify positive and negative behavior. For example, if you predict whether customers will buy a product, the outcomes where they buy it are positive; outcomes where they do not buy it are negative.

Extended scoring models

Extended scoring models predict a binary outcome and include outcome inferencing that is often used in risk management. Extended scoring models predict binary behavior with cases of unknown behavior. Cases can have unknown behavior when they are rejected by the business rules or system used to process their applications, or because they were accepted but subsequently the customer did not take up the offer. The prediction of the behavior of unknown behavior cases is based on the similarity with cases of known behavior and past acceptance business policies.

Note: You need a valid license to use the extended scoring functionality. Contact your account executive for licensing information.

Spectrum models

Spectrum models extend the concept of scoring models to the prediction of continuous behavior. Continuous behavior is a typically ordered range of values, for example, the number of items purchased or the length of a relationship.

A spectrum model calculates a score for each case and places it on a spectrum from the lowest to the highest value. The score range is divided into intervals where each interval is associated with the average value of the development sample cases that fall into the interval. This range provides the predicted value for new cases falling into each interval.

Behavior outside the score range is adjusted as if it had the maximum or minimum value of the range.