The SASSC Algorithm: an Application to Insurance Data

A classification technique is used to distinguish or discriminate between two (or more) groups of individuals. Frequently if the number of groups is large, the algorithm could have some problem to build a model that reaches a satisfying accuracy level. In order to overcome this problem we propose an approach called Sequential Automatic Search of a Subset of Classifiers (SASSC) that splits a classification problem among C classes into K C less complex two-classes sub-problems. SASSC focuses both on the interpretation aspects trying to improve the comprehension of the relationships among predictors, and on classification accuracy. The main contributions of SASSC concern the new criteria for the aggregation of classes and super-classes and the alternative criteria for the estimation of the response class for unseen (test-set) observation. An application of SASSC Algorithm to Insurance data is illustrated.