Evolutionary Computation for Assessing and Improving Classifier Performance



In this dissertation we explore the use of Evolutionary Computation for assessing and improving the performance of classifier systems, focusing on image classification tasks. A boosting framework for classifier improvement is proposed. The approach relies on the use of an evolutionary computation engine to exploit the potential weaknesses of the classifiers, evolving instances that produce classification errors. Subsequently these instances are become part of the training sets in order to circumvent the identified weaknesses. The framework was instantiated and tested in two image classification scenarios and several validation experiments were conducted. The experimental results attained are described and analyzed. Overall the results show the viability of the proposed approach and indicate future research.


Evolutionary Computation, Performance, Machine Learning, Examples Synthesis

MSc Thesis

Evolutionary Computation for Assessing and Improving Classifier Performance, September 2011

PDF File

Cited by

No citations found