Self-optimizing classifiers: formalization and design pattern

In this paper we propose a design pattern for self-optimizing classification systems, i.e. classifiers able to adapt their behavior to the system changes. First, we provide a formalization of a self-optimizing classifier we use to derive the design pattern. Then, we describe the pattern classes, their interactions, and validate our approach applying the proposed pattern to a real scenario. Finally, to evaluate the proposed solution we compare the behavior of the self-optimizing classifier with a not self-optimizing one. Experimental results demonstrate the approach effectiveness.