Comparative analysis of Mamdani and Sugeno inference systems for evaluating inter-agent dependency requirements

Abstract

The quantification and prediction of inter-agent dependency requirements is one of the main concerns in Agent Oriented Requirements Engineering. To quantify and envisage exertion load of an agent within resource constraints, this work provides a comparative analysis of Sugeno and Mamdani Inference Systems. The Mamdani Fuzzy Inference System (MFIS) goes well with the vagueness and ambiguity involved with inputs and output variables-whereas the Sugeno Fuzzy Inference System (SFIS) facilitates the linear relationship between inputs and output variables that is likely to assist the developer in tailoring the values of input domain attributes to obtain required level of DoD so as to develop Multi-Agent System (MAS) of high quality. The performance of these Inference Systems is examined by means of mean execution time. The results show that the SFIS outperforms MFIS. A sensitive analysis is employed to observe the behavior of MFIS and SFIS with respect to membership functions (MFs) viz. Guassian, Triangular and Trapezodial. The equivalence of the results is measured using performance indicators-Coefficient of Correlation (CORR) and the Normalized Root Mean Square Error (NRMSE). It is observed that MFIS in comparison to SFIS appears to be sensitive to the MFs. An adaptability analysis is augmented to illustrate the distinguished behaviors of the Inference Systems with respect to the frequent variations in input data. The results show that employing a SFIS could be a good option to quantify and customize dependency requirements in inter-agent communication.