Abstract
Many complex adaptive systems proposed models that attempt to utilize more than two problem solving tools or techniques such as fuzzy logic, machine learning, and genetic algorithms usually involve combining at least two techniques in one module, examples of such combinations are found in techniques such as machine learning using genetic algorithms, fuzzy machine learning, or fuzzy genetic algorithms. A tradeoff must be done between the combined technique's expected problem solving capability and between harvesting each individual technique's capability. We argue that, while integrating these methods may not significantly guarantee an increase of the ability of such systems in problem solving, but may also increase their complexity in a manner that represents a challenge for any optimization attempt. The narrow problem scope that these systems target also presents an objective that we attempted to address here. In this paper we proposed a novel algorithmic approach to optimize complex adaptive systems by emphasizing on their modularity property through segregating the used techniques into phases. We attempted to demonstrate the validity of our method by proposing a model consisting of four parts as follow: a fuzzy logic controller, a cluster-based adaptive genetic algorithm, an unsupervised machine learning algorithm, and the final component is a supervisory optimization algorithm that combines tuning modifiers of the parameters responsible for determining the overall results of the model's other three components. The model's resulting extreme complexity is due to its objective to cover a broad range of problem spaces and not pre-defined situations. We concluded that the modularity and adaptation of our presented model offers a promising and challenging unexploited territory of complex adaptive systems and their optimization attempts that require further exploration.