Estimating Software Logical Stability using ANN from Class diagram

Abstract

A problem that always occurs in software maintenance process is the ripple effect, which is when one object changed thus affecting the other objects in the same program. Stability can be defi ned as resistance to the ripple effect. This is divided into two aspects: logical stability and performance stability. In this paper, we emphasize the logical stability only. Logical stability of a class indicates its resistance to interclass propagation of changes when other classes are modifi ed. However, logical stability can be obtained only after the
program is executed. Thus, logical stability estimation at the design phase is very useful since knowing the amount of class logical stability may enable a designer to decide whether restructuring the design model is needed or not. In this paper, we propose a methodology to estimate logical stability of a class by using artifi cial neural network (ANN). Artifi cial neural network is an approach used to estimate the value of a class logical stability from historical data in the repository by choosing a multi layer Perceptron method. The establishment of this approach is evaluated for estimating the accuracy between the actual class logical stability measured from source code and an estimated class logical stability using ANN and Multiple Regression. The experiments result shows that Mean Magnitude Relation Error (MMRE) of ANN is 21.18 % and MMRE of Multiple Regression is 29.72 %. It shows that Artifi cial Neural Network is effective in the estimation.