A.F. Tulai and F. Oppacher (Canada)
Data Classification, Genentic Algorithms, speciation
In this paper we show improvements of accuracy in classification tasks, obtained in the framework of Evolutionary Computation. Different species of individuals are assigned to each class of objects simplifying the search for solutions. Each individual carries a short combination of attributes made from distinct features and the solution is represented by the whole population as the classification becomes a collective task. Pruning is not required. Using real-world learning tasks we compare the performance of our method with other classification SW packages (like C4.5) using traditional methods as well as newer methods like Support Vector Machines (SVMs) and Set Covering Machines (SCMs).
Important Links:
Go Back
