Abstract
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.