Improving Agent Coalitions by Behavioral Patterns Clustering and Conservative Reconfiguration

Abstract

In this paper we present a novel application of incremental data mining algorithms to the problem of formation and reconfiguration of coalitions of agents cooperating in dynamically evolving environment. Our experimental generator of coalitional structures takes into account both the stability of resulting coalitions and efficiency of computations. It focuses on providing nearly optimal solution in the average case and generates coherent stable groups with respect to agents beliefs, intentions, capabilities as well as the current environmental state. Incremental clustering leads to a robust adaptation of existing structure in response to rapidly changing environmental conditions. It aims at retaining the effectiveness of existing structure at the low reconfiguration costs