Bloat Free Genetic Programming versus Classification Trees for Identification of Burned Areas in Satellite Imagery
- Cite this paper as:
- Silva S., Vasconcelos M.J., Melo J.B. (2010) Bloat Free Genetic Programming versus Classification Trees for Identification of Burned Areas in Satellite Imagery. In: Di Chio C. et al. (eds) Applications of Evolutionary Computation. EvoApplications 2010. Lecture Notes in Computer Science, vol 6024. Springer, Berlin, Heidelberg
This paper compares Genetic Programming and Classification Trees on a problem of identification of burned areas in satellite imagery. Additionally, it studies how the most recently recognized bloat control technique, Operator Equalisation, affects the quality of the solutions provided by Genetic Programming. The merit of each approach is assessed not only by its classification accuracy, but also by the ability to predict the correctness of its own classifications, and the ability to provide solutions that are human readable and robust to data inaccuracies. The results reveal that both approaches achieve high accuracy with no overfitting, and that Genetic Programming can reveal some surprises and offer interesting advantages even on a simple problem so easily tackled by the popular Classification Trees. Operator Equalisation proved to be crucial.
Unable to display preview. Download preview PDF.