NeuroEAs-based algorithm portfolios for classification problems

Although an artificial neural network and evolutionary algorithms have been proved that they are efficient in many problems, the algorithms, generally, may produce good results with some problems and yield inferior solution in others. These cause risk of selecting an appropriate algorithm to solve a particular problem. This paper proposes a method to reduce risk of selecting an algorithm for solving classification problems by forming NeuroEAs-based algorithm portfolios to diversify risk. This method combines an artificial neural network and many different evolutionary algorithms to work together. It allocates existing computation time to the constituent algorithms, and encourages interaction among these algorithms consistently so that the algorithms can help improve performance of each other. The experiment results with 5 classification problems from UCI machine learning repository have shown that the algorithm portfolio outperforms its constituent algorithms given the same computation time.