Adaptive Neuro-Fuzzy Inference System for Assessing the Maintainability of the Software

Abstract

Measuring software maintainability at an earlier stage is a non-trivial task as it decides the software life cycle cost and customer satisfaction. Software designing is carried out using many object-oriented (OO) techniques. Among these, class modeling is one of the frequently used techniques. An enhanced Adaptive Neuro-Fuzzy Inference System (ANFIS) is proposed to assess the maintainability of the software at the design level. For measuring the maintainability, the metrics derived from the UML class diagram are used. The metrics namely coupling, and size are used as inputs for the proposed ANFIS based model. The size metric represents the structural complexity of the code whereas the coupling metrics represent the degree of interdependence between the software modules. The membership functions and the neural network parameters are determined based on the low mean square error value. The performance of the ANFIS model is evaluated using Root Mean Squared Error (RMSE), Coefficient of determination (R2) and Adj R2 techniques. Also, the performance of the proposed model is compared with Artificial Neural Network (ANN) model and the classical Fuzzy Inference System (FIS) model. The outcome of the ANFIS model reveals that it results in better performance when compared with ANN and FIS techniques.