Autonomous machines are interesting for both researchers and regular people. Everyone wants to have a self control machine that do the work by itself and deal with all types of problems. Thus, supervised learning and classification became important for high-dimensional and complex problems. However, classification algorithms only deals with discrete classes while practical and real-life applications contain continuous labels. Although several statistical techniques in machine learning were applied to solve this problem but they act as a black box and their actions are difficult to justify. Covering algorithms (CA), however, is one type of inductive learning that can be used to build a simple and powerful repository. Nevertheless, current CA approaches that deal with continuous classes are bias, non-updatable, overspecialized and sensitive to noise, or time consuming. Consequently, this paper proposes a novel non-discretization algorithm that deal with numeric classes while predicting discrete actions. It is a new version of RULES family called RULES-3C that learns interactively and transfer experience through exploiting the properties of reinforcement learning. This paper will investigate and assess the performance of RULES-3C with different practical cases and algorithms. Friedman test is also applied to rank RULES-3C performance and measure its significance.
CITATION STYLE
Elgibreen, H., & Aksoy, M. S. (2015). Classifying continuous classes with reinforcement learning rules. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 9012, pp. 116–127). Springer Verlag. https://doi.org/10.1007/978-3-319-15705-4_12
Mendeley helps you to discover research relevant for your work.