This study reviews the augmented Bayesian learning approach for the multi-agent decision mechanism. The decision mechanism includes the probability situation in a dynamic environment. Suggested system considers the problem of data analysis with distributed task allocation in a set of cooperative agents. This paper evaluates the exchange of information and the status of current information within the system and briefly examines the transition from knowledge base to rule base and from rule base to task, analyzing the sending of messages between agents when the information arrives on the system in a dynamic way. Both the current and past received information are taken into consideration together with related information rules are derived in the rule base structure by an augmented Bayesian classifier approach within knowledge base. In this paper, Expected Maximization (EM) algorithm, which is itself a learning method, has been developed and adapted for dynamic agent based data analysis.