Learning parameters of a genetic algorithm applied to signal classification

• Authors: A. J. Cantos;M. Santos
• Venue: AIA '08 Proceedings of the 26th IASTED International Conference on Artificial Intelligence and Applications
• Year: 2008

In this work, different parameters that have an influence on a signal classification system are presented. These parameters are related to both, the pre-processing task and the classification procedure. An evolutionary strategy focused on rule discovery is used to define the classification rules that are then applied to a database that consists of up to six different classes of fusion plasma signals. The influence of these control parameters on the final classification results are measured in terms of hits, misses and draws. The convergence and the learning rate of the genetic algorithm are also affected by some of these values. The evolutionary clustering is therefore optimized for this application. The results verify the effectiveness of the presented proposal. Some interesting conclusions can be derived.