Holonic agent-oriented Island Genetic Algorithms

Research areas in Holonic Multiagent Systems (HMAS) have found success in using selection-based methods, such as Evolutionary Algorithms (EAs), in order for agents to measure their own progress within the system, guided by a fitness function. There have also been approaches which apply Genetic Algorithms (GAs) to solving Supervised Learning (SL) problems. This research combines these existing approaches by introducing a holonic agent-oriented implementation of a recursively nested Island GA, called CLISDE, that builds a decision tree learner. The motivation for this work is to show that an agent-oriented implementation of the actor model is an appropriate framework for HMAS architectures. The unique contribution of this paper is that it expands the scope of MAS to the domain of SL and shows that when Island GAs are implemented in a holonic multiagent-based environment it results in a model expressing self-similarity of the problem type and the architecture, achieved by integrating the learning algorithm of decision trees directly into the evolutionary process. The CLISDE GA and three known parallel GA models, namely the Island GA, Master-Slave GA and Hierarchical GA, were evaluated on a classification dataset and their performance metrics were compared. The results indicate that the CLISDE GA produces a stronger decision tree predictor.