Rote-LCS learning classifier system for classification and prediction

Machine Learning (ML) involves the use of computer algorithms to solve for approximate solutions to problems with large, complex search spaces. Such problems have no known solution method, and search spaces too large to allow brute force search to be feasible. Evolutionary algorithms (EA) are a subset of machine learning algorithms which simulate fundamental concepts of evolution. EAs do not guarantee a perfect solution, but rather facilitate convergence to a solution of which the accuracy depends on a given EA\u27s learning architecture and the dynamics of the problem. Learning classifier systems (LCS) are algorithms comprising a subset of EAs. The Rote-LCS is a novel Pittsburgh-style LCS for supervised learning problems. The Rote models a solution space as a hyper-rectangle, where each independent variable represents a dimension. Rote rules are formed by binary trees with logical operators (decision trees) with relational hypotheses comprising the terminal nodes. In this representation, sub-rules (minor-hypotheses) are partitions on hyper-planes, and rules (major-hypotheses) are multidimensional partitions. The Rote-LCS has exhibited very high accuracy on classification problems, particularly Boolean problems, thus far. The Rote-LCS offers an additional attribute uncommon among machine learning algorithms - human readable solutions. Despite representing a multidimensional search space, Rote solutions may be graphed as two-dimensional trees. This makes the Rote-LCS a good candidate for supervised classification problems where insight is needed into the dynamics of a problem. Solutions generated by Rote-LCS could prospectively be used by scientists to form hypotheses regarding interactions between independent variables of a given problem. --Abstract, page iv