Analytical inference model for prediction and customization of inter-agent dependency requirements

Abstract

Inter-agent communication is one of the main concerns of Agent Oriented Requirements Engineering (AORE). The concern is delineated as managing inter-dependencies and interaction among various agents performing collaborative activities. To carry out cooperative activities, the application areas viz. electronic commerce and enterprise resource planning in the distributed environment require an agent to predict and customize dependency needs termed as Degree of Dependency (DoD) so that the goal may be obtained within resource constraints and with optimal number of agents. To quantify and predict exertion load of an agent within resource constraints, this paper proposes an Analytical Inference Model (AIM) that would facilitate the developer to evaluate and envisage DoD and hence analyze the optimum number of agents to obtain predicted DoD. In this work, Adaptive Neuro Fuzzy Inference System (ANFIS) combining the potential benefits of Artificial Neural Network (ANN) and Fuzzy Logic (FL) is employed to discover the linear relationship in input domain attributes and DoD. The resultant optimization of exertion loads would immensely improve the quality of the Multi-Agent System. The hybrid, as well as back propagation learning algorithm, is employed to adapt from training data. The bestfitness of proposed model against test data is examined by the performance indicators-Coefficient of Correlation (CORR) and the Normalized Root Mean Square Error (NRMSE). It is observed that hybrid learning algorithm outperforms the back propagation algorithm.