Comparative Analysis Of ANFIS And ANN For Evaluating Inter-Agent Dependency Requirements

The quantification and prediction of inter-agent dependency requirements is one of the main concerns in Agent Oriented Requirements Engineering. To evaluate exertion load of an agent within resource constraints, this work provides a comparative analysis of Adaptive Neuro Fuzzy Inference System (ANFIS) and Artificial Neural Network (ANN). ANN is widely known due to its capability of learning the training data while ANFIS is a hybrid system that combines the potential benefits of both the methodsFuzzy Inference System (FIS) and ANN. The performance analysis of these methods is accomplished using performance indicators-Coefficient of Correlation (CORR), the Normalized Root Mean Square Error (NRMSE) and Coefficient of Determination. It is observed that ANFIS results in highly correlated data points with least NRMSE over fitted with test data. Moreover decreasing rate of error in hybrid learning algorithm of ANFIS is found higher than back propagation of ANN learning algorithm. Mean execution time for both the methods is computed. Results show that the ANFIS outperforms ANN. Moreover ANFIS is equipped with the capability of generating a FIS with linear relationship in input-output data and hence facilitates analytical inference of fuzzy data, while ANN is limited to forecast the data using its learning potential. Hence employing ANFIS could be a good option to predict and customize dependency requirements in inter-agent communication. Keywords-component; Mamdani Fuzzy Inference System (MFIS), Sugeno Fuzzy Inference System (SFIS), Adaptive Neural Fuzzy Inference System (ANFIS), Artificial Neural Network (ANN), Degree of Dependency (DoD), Multi-Agent System (MAS).