An agent model for rough classifiers

This paper proposes a new agent-based approach in rough set classification theory. In data mining, the rough set technique is one classification technique. It generates rules from a large database and has mechanisms to handle noise and uncertainty in data. However, producing a rough classification model or rough classifier is computationally expensive, especially in its reduct computation phase: this is an NP-hard problem. These problems have brought about the generation of large amount of rules and high processing time. We solve these problems by embedding an agent-based algorithm within the rough modelling framework. In this study, the classifiers are based on creating agents within the main modelling processes such as reduct computation, rules generation and attribute projections. Four main agents are introduced: the interaction agent, weighted agent, reduction agent and default agent. We propose a heuristic for the default agent to control its searching activity. Experiments show that the proposed method significantly reduces the running time and the number of rules while maintaining the same classification accuracy.