Application of genetic programming classification in an industrial process resulting in greenhouse gas emission reductions

Abstract

This paper compares Genetic Programming and the Classification and Regression Trees algorithm as data driven modelling techniques on a case study in the ferrous metals and steel industry in South Africa. These industries are responsible for vast amounts of greenhouse gas production, and greenhouse gas emission reduction incentives exist that can fund these abatement technologies. Genetic Programming is used to derive pure classification rule sets, and to derive a regression model used for classification, and both these results are compared to the results obtained by decision trees, regarding accuracy and human interpretability. Considering the overall simplicity of the rule set obtained by Genetic Programming, and the fact that its accuracy was not surpassed by any of the other methods, we consider it to be the best approach, and highlight the advantages of using a rule based classification system. We conclude that Genetic Programming can potentially be used as a process model that reduces greenhouse gas production. © 2010 Springer-Verlag Berlin Heidelberg.