Chapter eleven - Analysis and explanation

Abstract

This chapter presents analysis and explanation tools that can be used to explain the performance of computational intelligence (CI) systems. Sensitivity analyses are important for determining how various inputs contribute to the output(s) of a system. They can be used during the system design phase to prune inputs that are irrelevant to the system output(s). Several approaches are employed for sensitivity analysis, including an approach featuring relation factors and a sensitivity analysis methodology. Explanation facilities are relatively common for traditional expert systems. Recently, using evolutionary computation tools such as genetic algorithms and particle swarm optimization (PSO) explanation facilities has been constructed for neural network systems. Hybrid diagnostic and classification systems incorporating a number of paradigms require explanation facilities that are useful, understandable, and consistent. These explanation facilities can be developed using a hybrid of rule-based and evolutionary computation tools. The chapter presents an example of an explanation facility for a neural network using the Iris dataset.